Arthur Calderon, as warden and on behalf of the State of California, hereinafter the “State,” appeals the district court’s partial grant of Thomas Martin Thompson’s petition for a writ of habeas corpus. The district court’s judgment set aside Thompson’s 1983 conviction in California state court for the rape of Ginger Fleischli. Thompson appeals the district court’s partial denial of the petition challenging his state court conviction for Fleischli’s murder and imposition of a sentence of death. The district court had jurisdiction under 28 U.S.C. § 2254. We have jurisdiction pursuant to 28 U.S.C. §§ 1291 and 2253. We reverse the partial grant of Thompson’s petition as to the rape charge. We affirm the district court’s partial denial of Thompson’s petition as to the murder conviction and penalty phase.
I
The evidence, both oral and physical, considered in the state trial court, the decisions made in state appellate proceedings and the matters considered in federal district court are reflected in the record before us.
A.
The state court trial record reveals the evidence presented for and against Thompson.
A California state court jury returned guilty verdicts against Thompson on one count of first-degree murder (Cal.Penal Code §§ 187, 189) and one count of forcible rape (Cal.Penal Code § 261(2)) in connection with the death of Ginger Fleischli. The jury also found true the special circumstance of murder during the commission of a rape (Cal.Penal Code § 190.2(a)(17)(iii)).
In a separate proceeding, David Leitch was found guilty of the second degree murder of Fleischli.
During the summer of 1981, Thompson occupied an apartment with David Leitch in Laguna Beach. Not long before, Ginger Fleischli, the 20-year-old victim, had moved out of the same apartment. Fleischli, who had been sexually involved with Leitch, moved in with Leitch’s ex-wife, Tracy Leitch.
On September 11, 1981, Fleischli, Tracy Leitch, David Leitch and Thompson met at a pizza parlor on Balboa Island. They removed to the Sandpiper Inn on Pacific Coast Highway in Laguna Beach, where Thompson and Fleischli remained drinking after David and Tracy Leitch departed. Afshin Kashani joined Fleischli and Thompson at about 9:30 p.m. The three drank and danced, both at the Sandpiper and at a nearby bar, the Boom-Boom Room. Kashani and Thompson also smoked some of Kashani’s hashish.
At about 1 a.m., Thompson, Kashani and Fleischli walked back to the apartment occupied by Thompson and Leitch. Kashani and Thompson smoked more hashish. Fleischli departed at about 2 a.m. to purchase some soda at a nearby liquor store. According to Kashani, Thompson indicated to him in Fleischli’s absence that he wanted to have sex with Fleischli that night and that Kashani could have her after Thompson left the country.1
At trial, Kashani testified that after he departed, he realized that he had forgotten his cigarettes and returned to the apartment. There, a nervous Thompson handed him the cigarettes through the door. Kashani looked *1363for Fleisehli at the liquor store; not seeing her, he went home.
Tracy Leitch testified that she went to the apartment the next morning and asked Thompson where Fleisehli was. She further testified that he told her that Fleisehli had left from the inn with Kashani.
Tracy Leitch again spoke with Thompson that evening about Fleischli’s whereabouts. She testified that Thompson answered using the past tense, saying that he had liked Fleisehli and that she was a nice girl. The next day, Tracy Leitch filed a missing person’s report with the police.
Fleischli’s body was found two days later in a grove of trees near an interstate highway. Two sets of footprints appeared near the body. One print matched a wavy soled shoe worn by David Leitch. Investigators could not identify the other footprint, which was smooth.
Fleisehli lay buried in dirt, wrapped in an old sleeping bag and a pink blanket that a previous tenant had left in the apartment occupied by Thompson and Leitch. Fibers from the blanket matched fibers found in David Leitch’s automobile. A stain on the rope used to tie Fleischli’s body matched paint from the trunk of Leitch’s vehicle.
Fleisehli was stabbed five times in the head near her right ear. A single-edged knife was thrust two and one half inches through Fleischli’s ear, penetrated her carotid artery, and caused massive bleeding and ultimately her death. Her head was wrapped with silver duet tape, two towels, a sheet and her jacket. Investigators found her blood in the carpet at the apartment occupied by Leitch and Thompson.
Fleisehli’s shirt and bra were cut in front and pulled down to her elbows. She wore no shoes, socks or underwear, and her jeans were fully zipped but not buttoned. A vaginal swab revealed semen consistent with Thompson’s blood type.
Before Fleischli’s body was discovered, Thompson and David Leitch travelled to Mexico together. Leitch returned and was arrested. Mexican authorities located Thompson, who possessed handcuffs at the time of his arrest. Thompson initially claimed that Fleisehli had left the apartment with Kashani. Thompson also told police that Fleisehli had been stabbed in the head, although the media had not previously reported this information. First denying that he had sexual intercourse with Fleisehli, Thompson later asserted that he and Fleischli had consensual intercourse in the apartment.
At the state trial, Dr. Richards, a forensic pathologist, testified that he found no anatomical evidence of rape. The pathologist expressed the opinion that several bruises on Fleischli’s body occurred at or near the time of her death.
A police investigator testified that one of Fleischli’s bruises resembled injuries he had seen and known to be caused by handcuffs.
Two jailhouse informants, John Del Frate and Edward Fink, testified that Thompson, while in jail, confessed to committing the rape and murder.
Thompson chose to testify despite his attorney’s advice to the contrary. Thompson testified that he had sex with Fleisehli and then passed out and slept until the next morning. He said that he heard nothing of the murder in the small apartment. He called witnesses who testified that he was a heavy sleeper, but he himself said that he had been awakened in Mexico by the sound of police cocking the hammers of the revolvers they pointed at his head.
The jury found Thompson guilty of rape and murder.
B.
After Thompson’s state appeals failed,2 Thompson filed a petition for a writ of habeas corpus in federal court.
*1364The district court conducted an evidentiary hearing on Thompson’s claim that his attorney, Ronald Brower, provided ineffective assistance of counsel. Thompson called Dr. Irving Root, a forensic pathologist who testified that he could have given rebuttal evidence in the state trial. A defense investigator contacted Root in 1983, but Brower did not call Root at trial. In 1983, the investigator gave Root sixteen documents, including the autopsy report, and asked for Root’s professional opinion about whether rape was likely and what kind of physical trauma is associated with rape. The district court restricted the State’s cross-examination of Root concerning his expertise and possible bias.
Brower testified at the evidentiary hearing. He defended his decisions, stating that he did not emphasize the rape charge because he based the defense on Thompson’s claim that he did not commit the murder. Brower testified that he had adequately rebutted the forensic evidence and that spending an inordinate amount of time on such evidence would detract from his strategy. Brower also testified that he believed that he conducted sufficient investigation and had adequately impeached the jailhouse informant testimony.
Both sides called attorney expert witnesses to support their arguments regarding ineffective assistance of counsel. One expert, Attorney James Stotler, characterized Brow-er’s representation as high caliber. Attorney Gerald Chaleff, the other expert, termed Brower’s representation ineffective and prejudicial to Thompson.
Following post-hearing briefing, the district court granted the petition in part. The court concluded that Brower’s failure to impeach Dr. Richards’ conclusions and to call a forensic expert, as well as his failure to impeach Fink with testimony concerning numerous prior instances where Fink had informed against others and received favors from law enforcement, rose to the level of ineffective assistance. The district court held that state trial counsel Brower prejudiced Thompson’s right to a fair trial on the rape charge. The district court also faulted Brower for failing to further impeach Del Frate. The district court held that the errors combined to produce prejudice. The State appeals this decision.
The district court rejected all of Thompson’s other challenges to the murder conviction and to the death penalty phase of the proceedings. Thompson appeals this decision.
II
We are mindful of the limited role of federal courts in habeas review of state convictions.
“The role of federal habeas proceedings, while important in assuring that constitutional rights are observed, is secondary and limited. Federal courts are not forums in which to relitigate state trials.” Brecht v. Abrahamson, 507 U.S. 619, 633, 113 S.Ct. 1710, 1719, 123 L.Ed.2d 353 (1993) (citation omitted).
We review de novo a district court’s order granting or denying a petition for a writ of habeas corpus. Sanders v. Ratelle, 21 F.3d 1446, 1451 (9th Cir.1994). The district court’s findings of fact are reviewed for clear error. Bonin v. Calderon, 59 F.3d 815, 823 (9th Cir.1995), cert. denied, — U.S. -, 116 S.Ct. 718, 133 L.Ed.2d 671 (1996).
This appeal presents the claimed ineffective assistance of defense counsel as a primary issue. A claim of ineffective assistance of counsel is a mixed question of law and fact which we review de novo. Harris v. Wood, 64 F.3d 1432, 1435 (9th Cir.1995) (citing Sanders, 21 F.3d at 1451). Both the performance and prejudice components of the ineffectiveness inquiry present mixed questions which we review de novo. Strickland v. Washington, 466 U.S. 668, 698, 104 S.Ct. 2052, 2070, 80 L.Ed.2d 674 (1984).
Thompson faces a “strong presumption” that counsel rendered adequate assistance and exercised reasonable professional judgment in making decisions. Id. at 690, 104 S.Ct. at 2065-66. Thompson must demonstrate that counsel’s performance was “outside the wide range of professionally competent assistance.” Id. Tactical decisions of trial counsel deserve deference *1365when: (1) counsel in fact bases trial conduct on strategic considerations; (2) counsel makes an informed decision based upon investigation; and (3) the decision appears reasonable under the circumstances. Sanders v. Ratelle, 21 F.3d at 1456.
Thompson must do more than show a deficiency of representation; he must also establish prejudice. He must demonstrate a “reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different. A reasonable probability is a probability sufficient to undermine confidence in the outcome.” Clabourne v. Lewis, 64 F.3d 1373, 1378 (9th Cir.1995) (quoting Strickland, 466 U.S. at 694, 104 S.Ct. at 2068). “The benchmark for judging any claim of ineffectiveness must be whether counsel’s conduct so undermined the proper functioning of the adversarial process that the trial cannot be relied on as having produced a just result.” Strickland, 466 U.S. at 686, 104 S.Ct. at 2064. Thompson bears the burden of showing that there is a reasonable probability that the jury would not have convicted him had defense counsel not rendered ineffective assistance.
Ill
The State appeals the district court’s partial grant of Thompson’s petition.
The district court concluded that Thompson’s trial counsel, Ronald Brower, rendered ineffective assistance in defending the rape charge. The court found the ineffective assistance cumulative and prejudicial, resulting from Brower’s failure to investigate, his failure to emphasize the lack of physical evidence of rape and his failure to investigate and more thoroughly cross-examine Fink and Del Frate.
A.
In finding prejudice, the district court minimized the evidence of rape and failed to account for the damage done by Thompson’s own testimony.
Brower testified in these proceedings that he chose not to emphasize the rape evidence in the state trial because he sought to prove that Thompson did not commit the murder. Thompson’s insistence on testifying that he did not rape Fleischli and slept through the murder played a significant part in Brower’s decision to adopt this strategy.
Because we disagree with the district court’s assessment of the strength of the State’s rape case, we confine our review to the prejudice element of the Strickland test. Cf. Williams v. Calderon, 52 F.3d 1465, 1470 (9th Cir.1995) (not reaching cause because petitioner had not shown prejudice), cert. denied, — U.S. -, 116 S.Ct. 937, 133 L.Ed.2d 863 (1996). The district court inexplicably found the evidence of rape weak but the evidence of murder “overwhelming,” even without the testimony of the jailhouse informants who testified to Thompson’s confession to both crimes. The State presented strong evidence of rape at the state trial. This evidence overlapped substantially with the evidence of murder. The jury convicted Thompson of both charges. The nature and amount of the evidence compel us to reverse the district court’s finding that Thompson had shown Brower’s alleged errors to have created a reasonable probability of a different verdict.
The State presented a strong rape case during the state trial. It demonstrated the existence of semen consistent with Thompson’s. On the question of consent, the condition of the body provided compelling evidence of rape. Someone had cut Fleischli’s shirt and bra up the middle and pulled them down to her elbows, placing her in a position consistent with restraint during a rape. Thompson suggested at trial that someone could have done this after killing her in order to facilitate moving her body, but the body was otherwise secured by the blanket, tape and ties. No logical inference supported exposing her breasts and restraining her in this fashion merely in order to move her dead body. Fleischli wore no shoes, socks or underwear, and her jeans had been fully zipped but not buttoned. She also suffered a severe crushing injury to her right wrist at or near the time of her death.
Most importantly, Thompson’s own testi*1366mony did severe damage to his defense.3 After having lied to police, Thompson admitted having intercourse with Fleischli that night in the apartment, and the jury did not have to believe Thompson’s testimony about being sound asleep while Leitch bound, gagged and killed the victim, all a few feet away. After the trial, the state court judge revealed the impact of Thompson’s fantastic claims when he took pains to indicate on the record Brower’s disagreement with Thompson’s decision to testify. The trial judge stated that he wanted to leave “no one believing that defense counsel joined in the folly of his client testifying in niy observation.” (Emphasis added). The district court agreed with this assessment of Thompson’s testimony, stating that it was “devastating to his defense.” The district court found it “riddled with inconsistencies and outright falsehoods,” and concluded that it “no doubt affected the jury’s verdict.”
Having reviewed the testimony, we would go further. As in Williams v. Calderon, 52 F.3d at 1470, we may here eschew an inquiry into Brower’s assistance in favor of a clear determination that Thompson has not shown prejudice, because his own words rank among the most damaging of the testimony presented to the jury. The district court concluded that Brower should have worked harder to rebut other evidence of rape and the testimony of the informants, but this would not have rebutted the condition and position of the bra and shirt on the body, the presence of some bruising and Thompson’s reluctant admission. After reviewing the opportunities that Thompson claims Brower missed, we cannot conclude that they would have made a difference in the verdict.
1.
First, we review the district court’s finding of ineffective assistance based upon Brower’s failure to rebut forensic evidence of rape.
Brower presented a defense theory in the state trial that Thompson had consensual sex and then Leitch bound, gagged and murdered Fleischli while Thompson slept. Counsel pursued this theory consistent with Thompson’s testimony, given against Brow-er’s advice, that he did not commit the murder or rape Fleischli. Brower stated that he did not want to emphasize the possibility of rape by spending an inordinate amount of time on the evidence of bruises. Instead, Brower sought to show that another person, David Leitch, caused the bruises and killed Fleischli while Thompson slept. The district court characterized Brower’s general strategy as “flawed,” and found his failure to systematically rebut each piece of medical evidence to constitute ineffective assistance of counsel.
a.
- We review the areas of forensic evidence that were the focus of the district court’s finding.
Prior to trial, Brower interviewed the State’s pathologist, Dr. Richards, who told Brower that he had found no “anatomical evidence” of rape. The parties vigorously dispute whether “anatomical evidence” meant vaginal bruising or any bruising. Brower testified that he asked Richards to clarify “anatomical evidence,” and Richards said that he had found no bruising, cuts or discolorations suggesting rape. Brower also testified that he and Richards went over Richards’ entire report, including the bruising Richards observed, and that Richards continued to repeat that the findings did not indicate rape. Brower felt that the State could not effectively impeach Richards on these points.
During his state court trial testimony, Richards noted bruises on the victim’s arms, wrists, palms and legs. He testified that the photographs of the victim’s body showed a “crushing injury with surrounding bruise to the right wrist.” A “large recent bruise on the left elbow” also appeared at the autopsy. He characterized a bruise to the right knee as older, but placed all the other bruises at or immediately before the time of death. He ascribed the bruising to “heavy handling,” *1367perhaps by hands causing pressure. He expressed the opinion that something heavy caused the crushing injury to the right wrist.
Having discussed the bruises, Richards concluded that there was no anatomical evidence of rape. Thompson vigorously disputes the meaning of the term “anatomical evidence” as used by Richards, arguing that it meant only vaginal bruising or tearing. Richards’ testimony supports the inference that by “anatomical evidence,” he meant any trauma, not just trauma to the vagina. On the meaning of the bruising, he testified that bruising to parts of the body other than the vagina usually demonstrated lack of consent even more than vaginal bruising. Just after making this statement, Richards admitted on reeross that, “[t]here is no anatomical evidence of rape. There is nothing to indicate it whatsoever. I would rely on the semen and sperm analysis.” Richards later clarified the term “anatomical evidence” by stating: “Most of the time there is no trauma sufficient to identify at autopsy ... there need be no evidence of injury at all with rape.” He characterized the determination of rape as “a legal matter ... out of [his] hands.” Richards’ 1988 testimony indicates that he found no physical evidence of rape aside from semen.
At the evidentiary hearing in 1993, Root testified that he would have disagreed with several of Richard’s conclusions.4 Root offered the opinion that the bruises on the victim’s legs occurred weeks before her death. Root ascribed the marks on the palms to lividity. He allowed that the contusion on the left hand could have occurred at the time of death, but he felt that the injury to the left wrist occurred a minimum of two days before death. He concluded that there was in fact no injury at all to the right elbow. As to the crushing injury to the right wrist, Root could not determine a time frame, but he testified that he saw no injuries consistent with his prior observations of injuries caused by handcuffs.
Deputy Darryl Coder had testified in 1983 that the injury to the right wrist looked like injuries he had seen inflicted by the locking mechanisms of handcuffs. Coder added that he could not say whether the injury occurred before or after Fleisehli was killed. The jury learned that Thompson had handcuffs in his possession when Mexican authorities arrested him. In 1993, Root specifically disagreed with the conclusion of Deputy Coder that handcuffs caused the injury to the right wrist.
The district court also focused on Richard’s statement that finding vaginal bruising or tearing caused by the forceful penetration of an erect penis was “rare as hen’s teeth.” Brower did not call an expert to rebut that testimony, but he did press Richards strongly on the point during cross-examination. Brower’s interrogation of Richards on cross-examination resulted in an admission that vaginal bruising or tearing was a common injury to rape victims. Richards also allowed that such injury represented a primary indication of rape as opposed to consensual intercourse. Brower questioned Richards on his insistence that this commonly found injury was only rarely caused by the forceful penetration of an erect penis, asking him if this meant that a large number of rape victims were penetrated by hard objects. While Richards stuck to his original statement, he stated that he always looked for ulceration or bruising as evidence of rape.
The logical conclusion from Brower’s cross-examination of Richards was that either Richards thought a large number of rape victims were penetrated by hard objects or he was mistaken regarding the improbability that an erect penis could cause vaginal bruising or tearing. Either way, Brower elicited the basic testimony that vaginal injury is common in rape cases and that Richards did not find it during his autopsy of the victim.
Finally, the district court noted Root’s testimony that the lack of semen in Fleisehli’s *1368jeans and the tightness of the jeans indicated that no rape had occurred. The district attorney impeached the latter theory on cross-examination.
b.
The alleged failure of Thompson’s trial counsel to rebut forensic evidence did not prejudice Thompson’s defense.
We do not reach the question whether Thompson demonstrated ineffective assistance because we find that he cannot show prejudice. Given Root’s 1993 testimony, Brower missed opportunities to rebut only certain aspects of the forensic evidence. In 1993 Root could not exclude the crushing injury to the wrist or the injury to the left hand. His testimony had no effect on the position in which the body was found, Thompson’s admission and the presence of Thompson’s semen.
As to the handcuffs, a jury in 1983 would have weighed Root’s expertise regarding handcuffs and injuries against Coder’s. Coder also qualified his testimony in important regards. Coder admitted that he could not determine whether the injury occurred before or after death, a key question for rape, and Root’s testimony regarding that injury would not have fixed it more clearly in time. Coder also admitted that he had only observed such injuries on living persons and that he had no formal training on the subject. The incremental addition of Root’s testimony would not have greatly altered the state of the evidence on this issue, which consisted of Coder’s opinion as qualified on cross-examination.
Most importantly, the State’s own pathologist remarked upon the lack of physical trauma and summarized the evidence by stating that he found no anatomical evidence of rape. Given the state of the evidence, we find that even if Dr. Root had been called to testify and had testified at the state court trial in the manner in which he did before the district court, there is not a reasonable probability of a different verdict.
We reverse the district court’s finding of prejudice relating to the physical evidence of rape.
2.
Second, we examine the district court’s determination that Brower rendered ineffective assistance by failing to sufficiently impeach the jailhouse informants, Fink and Del Frate.
a.
Fink testified that while housed in the same module as Thompson, Thompson told him about his crime. According to Fink, Thompson told him he “wanted to screw” Fleisehli, but she refused, so he “took it.” Fink testified that Thompson admitted to killing her “in a quiet way, like in the jungle,” so that she would not turn him in. Fink said that he came forward with the information while talking to investigators regarding another subject. He stated that Thompson’s killing of a woman upset him, although he allowed that the killing of a police officer would not have upset him in the same way.
Fink testified that he had asked to spend the remainder of his 29 days in prison in a different facility in return for his testimony. He admitted to four prior felony convictions and to recent heroin use. He added that heroin did not affect his brain. Fink admitted to using several aliases. Finally, Fink conceded that he informed on other inmates and that he stayed in protective custody due to this activity.
The district court listed information that Brower could have obtained about Fink. Fink had informed in several other murder cases, and he had received favors from police and prosecutors which had helped him win release on parole and lower sentences. Officials had also called Fink a liar and a confidence man. The district court found that, because Fink played a crucial role in the State’s rape case, Brower’s failure to discover and use Fink’s history of informing and receiving favors rose to the level of ineffective assistance. Reviewing the damaging evidence used by another defense lawyer against Fink, the district court made the factual finding that Brower could have obtained the information.
*1369The district court concluded that the jury would have disregarded Fink’s testimony if it had learned of the undiscovered information. This conclusion assumes that the jury credited Fink, and the district court based that assumption in part on its finding that other evidence of rape was lacking. As discussed above, however, the State put on a strong rape case even without the informant testimony. The record does not indicate that the jury must have believed Fink in order to convict Thompson of rape.
b.
The alleged shortcomings in state trial counsel’s impeachment of Fink did not prejudice Thompson.
Brower discredited Fink by showing that he received the special treatment of protected custody in return for informing, that he was known as a snitch, and that he asked for favors in return for his testimony. This demonstrated a strong motive to lie: Fink’s continued survival in prison. Brower mocked Fink’s statement that prisoners simply came up to him and told him things that he then told to police. Brower also called Steve Whitton, another prisoner, who testified that Fink obtained his information about Thompson from newspapers. We find no prejudice from Brower’s failure to impeach Fink further. Thompson has not demonstrated a reasonable probability of a different outcome had Brower further discredited Fink.
c.
Regarding Del Frate, the district court found that Brower had failed to discover certain impeachment evidence. The court termed most of the evidence cumulative, with the exception of Del Frate’s testimony at his 1978 murder trial regarding his long history as an informant. We note that this history of informing consisted largely of participating in drug buys and not of jailhouse informing. The court found that Brower erred by failing to locate and use this testimony, but determined that it did not independently create sufficient prejudice. The court concluded that it did add to the cumulative prejudice created by Brower’s other failures.
d.
No prejudice resulted from Brower’s failure to further impeach Del Frate.
Brower could hardly have impeached Del Frate more than he did. Brower brought out the fact that Del Frate did not initially tell investigators that Thompson admitted to forcing himself on Fleischli and stabbing her. Del Frate added these facts later, claiming that the investigators did not originally ask about them. Brower pressed Del Frate to explain inconsistencies in his statement about the stab wounds and the time frame for the rape. The jury was made aware of Del Frate’s felony convictions. Brower’s representation of Thompson as it concerned Del Frate did not prejudice Thompson.
e.
We conclude that Thompson’s trial counsel’s allegedly inadequate, impeachment of the jailhouse informants did not prejudice Thompson.
Brower impeached both jailhouse informants. Thompson argues that the State made much of their testimony during its closing, but the state court trial transcripts disclose that the informants were extensively cross-examined and discredited by Brower. We cannot say that the addition of the credited evidence, if any, that survived their impeachment formed a crucial part of the case.5
Thompson has not shown a reasonable probability of a different verdict were the informants impeached further.
B.
Thompson has not demonstrated errors sufficient to undermine confidence in the verdict. Finding no prejudice from the errors taken separately, we also find no cumulative prejudice. See Cooper v. Fitzharris, 586 F.2d 1325, 1333 (9th Cir.1978) (en banc) *1370(“prejudice may result from the cumulative impact of multiple deficiencies”), cert. denied, 440 U.S. 974, 99 S.Ct. 1542, 59 L.Ed.2d 793 (1979).
We reverse the district court’s granting in part of the petition on the rape charge.
IV
Thompson appeals the district court’s denial of the petition as to the murder conviction. Thompson challenges several of the district court’s conclusions.
A.
The district court held that the state trial court’s introduction of a hearsay statement by the victim and instructions to the jury regarding that statement did not violate Thompson’s rights.
On direct examination, Tracy Leitch testified that on the night she died Fleischli had asked “Do you think Tom [Thompson] would kill me?” On cross-examination, Leitch conceded that Fleischli had actually asked “Do you think David [Leitch] would have Tom kill me?”
At the state trial, the judge admitted the statement under California Evidence Code § 1250, as evidence of Fleischli’s state of mind indicating a lack of consent to the sexual intercourse with Thompson. The California Supreme Court found the statement properly admitted. Thompson, 246 Cal.Rptr. 245, 753 P.2d at 46-47. We do not review that particular determination. Estelle v. McGuire, 502 U.S. 62, 67, 112 S.Ct. 475, 479-80, 116 L.Ed.2d 385 (1991). Rather, we review Thompson’s claim that the state trial court’s instructions allowed the jury to consider the statement for the purpose of determining guilt of murder and thereby violated his constitutional rights.
The state trial court gave two instructions relating directly to the statement. Before the statement came into evidence, the court gave a limiting instruction to the jury:
You are not to consider the following testimony as proof of the truth that the murder occurred or that defendant committed such killing. You may consider it only for a determination of the state of mind of the victim, Ginger Fleischli, as it regards a rape and as it may negate a consensual sex act.
After repeating the last sentence for emphasis, the judge added, “That’s the only purpose for this, and it’s the only use you can put it to in your deliberations.”
In giving the final instructions to the jury, however, the court listed the statement as one of numerous factors that the jury could consider in determining guilt or innocence on the murder charge. This instruction directly contradicted the limiting instruction, and it was an error on the part of the trial court. Thompson, 246 Cal.Rptr. 245, 753 P.2d at 48-49. The California Supreme Court and the district court found this error harmless, however, due to the other instructions, the other evidence, and the closing arguments to the jury. See id.
Giving an instruction inadequate by direct appeal standards does not necessarily result in habeas relief from a state court conviction. Duckett v. Godinez, 67 F.3d 734, 745 (9th Cir.1995), cert. denied, — U.S. -, 116 S.Ct. 1549, 134 L.Ed.2d 651 (1996). The error must rise to a constitutional violation, which depends “upon the evidence in the case and the overall instructions given to the jury.” Id. (citing Cupp v. Naughten, 414 U.S. 141, 146, 94 S.Ct. 396, 400, 38 L.Ed.2d 368 (1973)).
Taken as a whole, the jury instructions did not highlight the statement as a crucial factor for consideration on the murder charge. In addition to giving the limiting instruction during the testimony, the trial court instructed the jury at the end of the trial that it could also consider the following in determining guilt on the murder charge:
Seven, that on no occasion prior to the victim’s demise had the defendant expressed any animosity, ill will or intent to harm the victim. Eight, that prior to the victim’s demise she had expressed a fear of physical harm from David Leitch. Nine, that prior to the victim’s demise David Leitch had expressed hate and ill will toward the victim and had threatened to have her killed.
*1371(Emphasis added). These factors, by implication, reduced the impact of the erroneous seventh factor given in the other instruction. In addition, the judge reminded the jury of his previous limiting instructions and cautioned the jury not to single out any particular sentence or instruction and ignore all the others. Taking the jury instructions as a whole, it is unlikely that the jurors singled out the erroneous instruction over and above the limiting instruction, the other factors listed, and the instruction reinforcing the limiting instruction.
The prosecutor’s references to the statement in rebuttal argument also weigh in favor of finding harmless error. The prosecutor mentioned the statement three times in rebuttal, but he never singled the statement out as evidence of murder as opposed to evidence of rape.
Finally, the evidence of murder against Thompson was compelling. Listed as it was among other stronger factors weighing in favor of guilt, Fleischli’s statement could not have played a critical role in the verdict. If the statement as amended by Tracy Leitch on cross examination indicated anything crucial to murder, it showed a fear of David Leitch. Because the jury instructions as a whole and the evidence do not indicate that the trial became fundamentally unfair due to the erroneous instruction, we affirm the district court on this ground.
B.
The district court rejected Thompson’s claim that the State violated his due process rights by pursuing inconsistent theories at his trial and at the trial of David Leitch.
Few courts have addressed this argument, but the principles it invokes are well established. Due process requires a fair trial in a fair tribunal. Turner v. Louisiana, 379 U.S. 466, 85 S.Ct. 546, 13 L.Ed.2d 424 (1965). The prosecutor represents the State, and that representation entails not only attention to the interests of the victim and survivors, but also to society’s interest in the fairness of criminal justice administration. Brady v. Maryland, 373 U.S. 83, 86, 83 S.Ct. 1194, 1196, 10 L.Ed.2d 215 (1963). A prosecutor may not treat a defendant in a manner that undermines society’s confidence in the fairness of the process. Id. For example, the prosecutor may not obtain a conviction through the use of false evidence. Mooney v. Holohan, 294 U.S. 103, 112, 55 S.Ct. 340, 341-42, 79 L.Ed. 791 (1935); Napue v. Illinois, 360 U.S. 264, 269, 79 S.Ct. 1173, 1177, 3 L.Ed.2d 1217 (1959). From these principles emerges the requirement that the prosecutor not pursue wholly inconsistent theories of a case at separate trials. See Haynes v. Cupp, 827 F.2d 435, 439 (9th Cir.1987) (variations in state’s theories not cause for reversal where theme that each defendant was culpable remained consistent); Drake v. Francis, 727 F.2d 990, 994 (11th Cir.1984), rev’d in part on other grounds, Drake v. Kemp, 762 F.2d 1449 (1985) (en banc), cert. denied, 478 U.S. 1020, 106 S.Ct. 3333, 92 L.Ed.2d 738 (1986).
Presenting evidence about Fleischli’s death, the State emphasized Thompson’s involvement at Thompson’s trial, and emphasized David Leitch’s involvement at Leiteh’s trial. The approach at Thompson’s trial included the prosecutor’s unsuccessful objection on relevance grounds to evidence of Leitch’s hostility toward Fleischli and argument by the prosecutor that only Thompson was in the apartment with Fleischli when she was killed. In downplaying Leitch’s role, the State never argued that the witnesses testifying to Leitch’s hostility lacked credibility. This fact deflates Thompson’s claim that the State presented factually inconsistent evidence when it called these same witnesses at Leiteh’s trial. The prosecutor had never indicated that the jury should disbelieve these witnesses; he had simply argued that the evidence in Thompson’s trial pointed to Thompson as the killer, notwithstanding Leitch’s hostility. We find no prejudice resulting from this theory.
Further, we conclude that if any prejudice had resulted from the differences in emphasis, it would have affected Leitch, not Thompson. See Drake v. Francis, 727 F.2d at 994 n. 4 (noting that any possible prejudice hurt the defendant not before the court). The trial records in both cases indicate that the State placed Thompson in the apartment at the time of the murder, and the State *1372never argued that Thompson slept through the murder or otherwise failed to participate. Because the prosecution’s theory in Leitch’s trial did not in any way exonerate Thompson, we affirm the district court’s disposition of this issue.
C.
Because trial counsel’s alleged ineffective assistance in defending the rape charge did not prejudice the defense of that charge, this alleged ineffective assistance also did not prejudice the defense of the murder charge.
Thompson contends that Brower’s alleged failings in defending the rape charge also prejudiced his defense of the murder charge. Thompson argues that the jury, by finding him guilty of rape and murder, accepted the prosecution’s felony murder theory, and therefore the ineffective assistance on the rape charge also undermines the murder charge.
We agree that the case presented by the State did not allow for a clear division between evidence of rape and murder. We determine, however, that because Brower’s conduct did not prejudice Thompson’s defense of the rape charge, it does not undermine confidence in his murder conviction. Because of the strong evidence of rape, we find no “reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Clabourne v. Lewis, 64 F.3d at 1378 (quoting Strickland, 466 U.S. at 694, 104 S.Ct. at 2068).
We affirm the district court’s denial in part of the petition on the murder charge.
V
Thompson appeals the district court’s denial in part of the petition as to the penalty phase of the state court proceedings.
The facts of the penalty phase are fully set out in People v. Thompson, 246 Cal.Rptr. 245, 753 P.2d at 58-59 and in the district court order. The State presented no new evidence. Thompson presented the testimony of family and friends regarding Thompson’s childhood and his good qualities. Two psychologists, Dr. Kara Cross and Dr. Francis Crinella, also testified for Thompson, describing the personality disorders he suffered as the result of a childhood spent living with his strict stepfather. After hearing this evidence and arguments from counsel, the jury returned a death verdict. Thompson raises several challenges to the penalty stage of the proceedings.
A.
Thompson argues that his state trial counsel’s ineffective assistance at the guilt phase prejudiced the penalty phase. Thompson has not carried his burden of demonstrating sufficient prejudice at the guilt phase to create a reasonable probability that the jury would have reached a different verdict. Likewise, we determine that Thompson has similarly failed to show that these alleged failings of defense counsel prejudiced the penalty phase.
B.
Thompson challenges the district court’s conclusion that the prosecutor’s reference in closing argument to an invalid aggravating factor did not prejudice the penalty phase.
At the close of the penalty phase, the prosecutor told the jury:
The only other evidence that we have presented in the guilt phase about other acts of violence would be the defendant’s plan to kill anyone who got in his way or the testimony we’ve heard from Mr. Del Frate about the defendant’s plan to kill the co-defendant, Mr. Leitch.
Thompson contends that this argument prejudiced him by adding an aggravating factor of other criminal conduct and negating the mitigating factor of his clean criminal record.
The California Supreme Court found the prosecutor’s argument invalid under Cal.Penal Code § 190.3(b), which requires proof beyond a reasonable doubt of conduct constituting commission of an actual crime. People v. Thompson, 246 Cal.Rptr. 245, 753 P.2d at 62 (citation omitted). However, both the California Supreme Court and the district court found no prejudice resulting from this *1373argument. Id. 246 Cal.Rptr. at 282, 753 P.2d at 64 (finding it “harmless under any standard----”)• We agree.
The arguments of counsel are generally accorded less weight by the jury than the court’s instructions. Boyde v. California, 494 U.S. 370, 384, 110 S.Ct. 1190, 1200, 108 L.Ed.2d 316 (1989). The arguments of counsel must be judged in the context of the entire argument and of the instructions given by the court. Id. at 385, 110 S.Ct. at 1200-01. The prosecutor mentioned the challenged portion of the argument only once and did not emphasize it. The court instructed the jury that any set of facts considered as an aggravating circumstance must be found by all the jurors to be proven beyond a reasonable doubt. We assume that the jury followed this instruction, and Thompson has shown no reasonable probability that the jury found any other crime proven beyond a reasonable doubt.
Thompson relies on People v. Lucero, 44 Cal.3d 1006, 245 Cal.Rptr. 185, 750 P.2d 1342, 1349 n. 5 (1988), where the prosecutor had argued that the circumstances of the crime charged also qualified as a crime of violence under Cal.Penal Code § 190.3(b) and that the absence of evidence of certain mitigating factors rendered these mitigating factors aggravating. The court noted that the combination of these improper arguments “turned the case upside down,” leaving the jury with several aggravating factors and only one mitigating factor: the absence of felony convictions. Id. This portion of Lucero is dicta; the court stated that it did not base its reversal on the improper arguments and did not assess their prejudicial effect. 245 Cal. Rptr. at 201 n. 17, 750 P.2d at 1357 n. 17. Moreover, Lucero involved two problematic arguments, one doubling the aggravating impact of the crime charged and another specifically turning the absence of mitigating factors into the presence of aggravating ones. The prosecutor’s solitary argument here does not present a Lucero problem.
The jury could not have concluded that any other crimes were proven beyond a reasonable doubt. Thompson’s lack of criminal history and lack of other criminal conduct stood as a mitigating factor. The jury was free under the state trial court’s instructions to find that the mitigating circumstances were outweighed by the crimes of which Thompson was convicted.
C.
Thompson challenges several of the state trial court’s decisions regarding jury instructions at the penalty phase.
A faulty jury instruction compels the issuance of a writ only where the instruction so infects the entire trial that the resulting conviction violates due process. Cupp v. Naughten, 414 U.S. at 147, 94 S.Ct. at 400-401. Where the habeas challenge rests on the failure to give an instruction, the already weighty burden on a petitioner becomes “especially heavy.” Henderson v. Kibbe, 431 U.S. 145, 155, 97 S.Ct. 1730, 1737, 52 L.Ed.2d 203 (1977).
Thompson argues that the penalty instruction misled the jury into believing that it must impose a death penalty if aggravating factors outweighed mitigating factors. The Supreme Court addressed and rejected this argument as applied to the same instruction in Boyde v. California, 494 U.S. at 377, 110 S.Ct. at 1196. We reject it here.
Thompson also contends that the California Supreme Court’s decisions finding the instruction invalid result in an unconstitutional differential application of the law in his case. We reject this argument because Thompson failed .to raise it before the district court.
Thompson contends that the court erred in declining his jury instruction on potential for rehabilitation and that the prosecutor improperly argued that he might kill again. Thompson presented no evidence on the specific issue of his potential for rehabilitation.
Moreover, Thompson makes no showing that the jury was precluded from considering his potential for rehabilitation despite the omission of the instruction. While the court rejected Thompson’s lengthy proposed instruction that listed rehabilitation as a mitigating factor, the court instructed the jury that: “The mitigating circumstances *1374which I have read for your consideration are given to you merely as examples of some of the factors which you may take into account____” Thompson does not satisfy his heavy burden of showing that the omission of a more specific instruction violated his due process rights.
The prosecutor’s single reference to the chance that Thompson might kill again in prison did not prejudice Thompson. This argument stood alone, and the jury was instructed not to speculate or give weight to mere possibilities.
D.
Thompson argues that his state trial counsel rendered ineffective assistance at the penalty phase.
Thompson claims that Brower failed to investigate his background and to interview the psychologists, resulting in the presentation of some damaging testimony. Noting that Thompson did not raise the issue at the evidentiary hearing, the district court found that Brower made an informed tactical decision to put on the psychologists. The court also noted that Dr. Crinella told the defense investigators that his testimony would be damaging. We agree with the district court that Brower’s decision to put on the testimony for its helpful aspects deserves deference under Strickland as a tactical decision made at the time of the state trial.
Thompson also argues that Brower had no coherent penalty phase strategy, pointing to inconsistencies between Brower’s closing argument, where he conceded guilt, and a requested instruction allowing the jury to consider Thompson’s lack of intent to kill. We do not find the argument itself to constitute ineffective assistance under the deferential Strickland standard. Brower was present at the trial. He saw the evidence and its effect on the jury. At the penalty phase a defense attorney may make the tactical decision to indicate that he identifies with the thought processes of the jurors. . We find no possible prejudice from the inconsistency of the argument and the requested jury instruction.
Thompson argues that Brower’s failure to request proper instructions on mitigation, the effect of a death verdict and lingering doubts constituted ineffective assistance. Considering the given instructions as a whole, we- find that Thompson has not shown that Brower’s failure to frame better instructions and the court’s failure to give instructions in these areas does not undermine confidence in the outcome.
Having carefully reviewed the record of the penalty phase of the state court trial proceedings, we find no cumulative prejudice sufficient to create a reasonable possibility of a different result.
VI
We conclude that Thompson has not demonstrated prejudice resulting from the alleged ineffective assistance of trial counsel.
Although Thompson raises numerous claims of ineffective assistance at the guilt and penalty phases, he has not shown a reasonable probability that the incremental addition of certain evidence or the incremental reduction in the effect of other evidence would have resulted in a different verdict. The State’s case against Thompson was strong on both the rape and the murder charges. Thompson himself made it much stronger by testifying after counsel advised him not to testify. We are satisfied that the jury would not have reached a different outcome either at the guilt or penalty phase had counsel done all the things that Thompson now claims he should have done. We are confident that the jury reached a just result in a fair proceeding.
We reverse the district court’s granting in part of the petition as to the rape conviction. We affirm the district court’s denial in part of the petition as to the murder conviction and the death verdict. Because we reverse the district court’s granting in part of the petition, the state conviction and sentence ■will not be disturbed and a writ will not issue.
REVERSED in part and AFFIRMED in part.
. Thompson apparently planned to leave the country in order to smuggle Southeast Asian refugees, gold, and possibly cocaine out of Thailand. The State suggested at trial that Fleischli's knowledge of this plan formed part of Thompson’s motive for murder.
. The California Supreme Court unanimously affirmed his convictions for rape and murder. People v. Thompson, 45 Cal.3d 86, 246 Cal.Rptr. 245, 753 P.2d 37 (1988), cert. denied, 488 U.S. 960, 109 S.Ct. 404, 102 L.Ed.2d 392 (1988). State collateral proceedings did not relieve Thompson from the state court judgment and sentence. People v. Thompson, No. S008382 (1989); In re Thompson, No. S018049 (1991).
. Thompson does not raise on appeal his claim of ineffective assistance based on Brower’s alleged failure to adequately prepare him for testifying or dissuade him from testifying. Thompson himself admitted that he ignored Brower's advice.
. Because we find no prejudice, we do not reach the State’s claims of clear error in the district court’s credibility determinations and of abuse of discretion in limiting the State’s cross-examination of Root on the issues of his bias and expert credentials. For the purpose of this section, we assume Root’s qualifications and lack of bias and we do not rely upon Brower’s explanations of his conduct.
. We note again the district court’s conclusion that the informants formed a crucial part of the rape case while the same alleged failings in Brower’s impeachment of them did not affect the murder case,