We consider here whether a pension plan that excludes a high number of participants from vesting violates the Labor Management Relations Act (LMRA). In particular, we consider whether the maintenance of restrictive vesting standards amounts to arbitrary and capricious conduct or creates a structural defect under § 302(c)(5) of the LMRA, 29 U.S.C. § 186(c)(5).
The Alaska Hotel and Restaurant Employees Pension Fund (the Fund) appeals a judgment that it violated the LMRA by failing to adopt vesting standards that would have allowed a greater number of participants to obtain benefits. Many participants whose pensions did not vest were former employees on the Trans-Alaska pipeline.
The Fund makes two arguments supporting reversal: (1) the district court erred in concluding that the plan contained a structural defect and that the trustees acted arbitrarily and capriciously, and (2) the plaintiffs’ class action is time barred under the applicable statute of limitation. Finding no violation of the LMRA, we reverse the judgment of the district court.
BACKGROUND
The Fund was established in 1966 pursuant to an agreement between two local chapters of the International Union of Hotel and Restaurant Employees and the Association of Alaska Hotel and Restaurant *512Owners. It is administered by a board of trustees comprised of an equal number of employer and union representatives. The trustees determine pension eligibility requirements and benefit levels.
The Fund was set up as a defined benefit plan. This type of plan pools contributions rather than segregating them in individual accounts. Employers make contributions on behalf of all employees regardless of the length of their employment. Participants receive specific benefits at retirement based on a formula focusing on years of participation in the plan. Nonvested participants do not receive benefits.
An employee’s pension benefits vest when he or she has attained a certain number of hours or years of credited service. Over the years, the Fund has refined its vesting requirements. Its rules have at all times complied with the minimum vesting standards required under the Employee Retirement Income Security Act of 1974 (ERISA).
From 1973 to 1986, ten years of credited service were required before pension rights vested. Before 1976, a participant was required to work at least 500 hours in a calendar year to obtain one year of credited service; in 1976, this minimum was reduced to 435 hours. Effective May 1,1976, and continuing to the present, an alternative provision enables a participant to vest after 15,000 hours of credited service, regardless of his or her years of credited service. Vesting was further liberalized after this action was filed: the trustees amended the plan to grant benefits to active participants with five or more years of credited service as of May 1, 1986.
The culinary trade in Alaska provides jobs in two very different environments: “town” jobs in the hotels and restaurants of populated areas, and “camp” jobs in the dining halls of remote construction sites. Workers in both types of jobs are participants in the Fund.
During construction of the Trans-Alaska pipeline, the number of camp employees increased markedly. The surge in pipeline-related employment nearly trebled the total number of plan participants between 1973-74 and 1976-77, from 2,791 to 7,004. When pipeline construction ended in 1978-79, the number of participants decreased to 2,683.
This sharp drop represented primarily pipeline workers who had less than three years of service. Many of these former camp employees could not find employment after the pipeline was completed that would have allowed them to remain covered by the Fund. The Fund has always had a “break-in-service” rule under which a participant forfeits all years and hours of credited service if he or she fails to work the annual minimum number of hours (435 hours since 1976) over a given number of years. The district court found that many members of the plaintiff class suffered involuntary breaks in service when their jobs ended and they were unable to find employment covered by the Fund.
The exodus of large numbers of pipeline workers with relatively few years of service increased the number of former participants whose benefits never vested. As of 1970-71, 4.28 percent of Fund participants ultimately qualified for a pension benefit. By 1979-80, this percentage had dropped to 1.31. When this lawsuit was initiated in 1986, 2.67 percent of the Fund’s 58,042 participants had qualified for benefits. The Fund’s “exclusion rate” was thus 97.33 percent.
The Fund reached “full funding” in 1979, 1986, 1987 and 1988. Full funding exists when plan assets exceed the sum of its actuarial liabilities and the year’s expected costs. Full funding may result in the loss of income tax deductibility for employer contributions and the imposition of an excise tax. The trustees voted to increase benefits to those already eligible in 1975, 1976, and 1980. They provided additional increases after this suit was filed.
Three named plaintiffs brought this class action in 1986 charging that the Fund’s maintenance of restrictive vesting rules violated both § 302(c)(5) of the LMRA and § 404 of ERISA. The district court certified a plaintiff class consisting of all non-vested participants on whose behalf contributions were made to the Fund after January 1, 1973, and who had three or more *513years of credited service, exclusive of any breaks in service of less than five years.1
Evidence was presented at trial that contributions to the Fund made on behalf of camp workers averaged $2.00 or more per hour from 1974-86. During this same period, contributions for town workers averaged $.30 to $.55 per hour. The district court found that class members were responsible for contributions of approximately $23,900,000 ($47,000,000 with interest).
The district court concluded that as early as 1974, the trustees knew or should have known that the pipeline project would dramatically increase the Fund’s population, and that by 1977 many of the new participants would lose their camp jobs and probably forfeit their credited service under the Fund’s then-existing rules. Focusing on the exclusion rate and the trustees’ decisions to increase benefits rather than relax eligibility rules, the court found the trustees’ conduct arbitrary and capricious. It held that maintenance of the 10-year/15, 000-hour requirement was a structural defect that violated the LMRA. It concluded that neither the reduction of the annual service requirement to 435 hours nor the adoption of the 15,000-hour alternative materially improved the exclusion rate.
The court also examined the income-replacement ratio of those employees who ultimately received pensions. This ratio compares a retiree’s pension income plus Social Security to his or her net income while last employed. The court found a LMRA violation partly because it concluded that a “significant number” of fund retirees receive benefits exceeding 100 percent of what they last earned. The court’s holding was also based on the Fund’s lack of reciprocity2 or self-pay3 provisions.
The court’s remedy was premised on its conclusion that since at least 1979 the trustees owed class members the duty of easing eligibility rules. To remedy the structural defect, the court revised the Fund’s eligibility rules, ordering five-year vesting retroactive to January 1, 1973. It also required the Fund to offer a self-pay option that would permit workers with four years of service to pay for the additional year needed to vest. Although the revisions were to be retroactive, the court did not award plaintiff class members past unpaid benefits. It simply granted full vesting from the date of judgment to those class members who qualified under the new rules.
DISCUSSION
Before turning to the merits of the case and the statute of limitation issue, we address the Fund’s argument that the court lacks subject matter jurisdiction over the LMRA and ERISA claims.
I. Subject Matter Jurisdiction
The Fund argues that a federal court lacks subject matter jurisdiction over an action to prove a structural defect under the LMRA when there is no evidence that the plan violated ERISA’s minimum vesting standards. For that reason, it also contends that the court lacks jurisdiction over plaintiffs’ claim that the trustees breached their fiduciary duties under ERISA.
Although the Supreme Court has yet to address the issue directly, see United Mine Workers Health & Retirement Funds v. Robinson, 455 U.S. 562, 573 n. 12, 102 S.Ct. 1226, 1233 n. 12, 71 L.Ed.2d 419 (1982), this circuit has long assumed jurisdiction over actions premised on the LMRA’s “sole and exclusive benefit” provision. We are not divested of such jurisdiction simply because a plan has complied *514with minimum ERISA vesting standards. See Hurn v. Retirement Fund Trust of Plumbing, Heating & Piping Indus., 703 F.2d 386, 391 (9th Cir.1983).
The Fund and amicus National Coordinating Committee for Multiemployer Plans (NCCMP) also argue that ERISA preempted or impliedly repealed § 302(c)(5) of the LMRA. We disagree. In enacting ERISA, Congress debated extensively the issue of vesting requirements before settling on a ten-year minimum for multiemployer plans.4 It set the minimum at ten years, rather than a shorter time, partly because the additional expense of a shorter minimum would impede the adoption of new plans and the liberalization of existing ones. H.R.Rep. No. 93-807, 93d Cong., 2d Sess., reprinted in 1974 U.S.Code Cong. & Admin.News 4670, 4685-86. There is also evidence to suggest that Congress rejected a shorter minimum because it could undermine the interest of employers in retaining long-term employees.5
It does not necessarily follow, however, that in enacting a statutory minimum, Congress also intended to preempt actions under § 302(c)(5) of the LMRA where a plan complies with the ten-year requirement. In rejecting a nearly identical argument, we said:
The Fund’s broader argument that ERISA preempts the [LMRA] on pension issues is equally untenable. Both the Supreme Court and this court have recognized that the [LMRA] provisions parallel the ERISA provisions and that trustees must meet the requirements of each. Where Congress intended ERISA to repeal or supersede other laws, state or federal, it said so. But it said nothing about section 302(c)(5). ERISA was not to affect any federal laws not specifically mentioned, and it does not preempt Hum’s [LMRA] claim.
Hum, 703 F.2d at 391 (citations omitted).
Because ERISA essentially codified the fiduciary standards of the LMRA, see NLRB v. Amax Coal Co., 453 U.S. 322, 332, 101 S.Ct. 2789, 2795, 69 L.Ed.2d 672 (1981), the ERISA ten-year minimum is most reasonably interpreted as just that: a statutory minimum that is intended to coexist with the fiduciary duties imposed by both ERISA and the LMRA. A shorter period may sometimes be appropriate and necessary to insure that eligibility rules operate for the “sole and exclusive benefit” of employees.
Similarly, a shorter vesting period may be appropriate and necessary to insure that the trustees fulfill their fiduciary duties under ERISA. See 29 U.S.C. § 1104(a)(l)(A)(i) (1988) (trustees must act “solely in the interest of the participants and beneficiaries and — for the exclusive purpose of providing benefits to participants and their beneficiaries ... ”). As Senator Williams stated: “[ERISA’s] provisions are only minimum standards, and no employer is impeded from building upon or improving these minimum requirements.” 2 Legislative History of ERISA, supra note 5, at 1601.
We hold that a structural defect action may be maintained under § 302(c)(5) of the LMRA and a breach of fiduciary duty claim may be maintained under § 404 of ERISA notwithstanding a plan’s compliance with minimum ERISA vesting requirements. This does not make ERISA compliance irrelevant. The court may consider *515such compliance as a factor in determining whether the trustees’ conduct was arbitrary and capricious or whether the plan was structurally defective.
II. Structural Defect Analysis
A. Standard of Review
“Courts are extremely reluctant to substitute their judgments for the judgments of trustees and will do so only if the actions of the trustees are not grounded on any reasonable basis.” Ponce v. Construction Laborers Pension Trust, 628 F.2d 537, 542 (9th Cir.1980) {Ponce I). Short of “plainly unjust” standards, trustees have wide latitude in formulating pension eligibility rules. Id. at 544. “A court should interfere only when the rule is unreasonable or its enforcement arbitrary.” Giler v. Board of Trustees, 509 F.2d 848, 849 (9th Cir.1974).
The questions whether the trustees have acted arbitrarily and capriciously or whether the plan contains a structural defect are mixed ones of law and fact. When reviewing the ultimate questions of arbitrary and capricious conduct and structural defect, this court has not accorded any special deference to the district court’s determinations. See, e.g., Elser v. I.A.M. Nat’l Pension Fund, 684 F.2d 648, 654 (9th Cir.1982). We review de novo these ultimate conclusions. Underlying findings of fact are reviewed for clear error. See Miranda v. Audia, 681 F.2d 1124, 1127 (9th Cir.1982), cert. denied, 464 U.S. 813, 104 S.Ct. 67, 78 L.Ed.2d 82 (1983).
B. Application of § 302(c)(5)
A pension plan is structurally deficient when it arbitrarily and unreasonably excludes a large number of participants from receiving benefits, thus failing to satisfy the § 302(c)(5) requirement that it be for the “sole and exclusive benefit” of all employees. Ponce I, 628 F.2d at 543. The plaintiff carries the initial burden of showing that the plan excludes an “unusually high” number of participants. Id. If the plaintiff does so, the burden shifts to the trustees to show the reasonableness of the plan’s eligibility requirements. Id. The district court’s finding that the exclusion rate was excessively high is a finding of fact reviewed for clear error. Ponce v. Construction Laborers Pension Trust, 774 F.2d 1401, 1402 (9th Cir.1985) (per curiam) (Ponce III), cert. denied, 479 U.S. 890, 107 S.Ct. 289, 93 L.Ed.2d 263 (1986).
Ultimately, we must decide whether the Fund’s vesting standards were so arbitrary and capricious that they amounted to a structural deficiency. The district court’s reasoning and remedy require us to consider the combined effects of the Fund’s 10-year/15,000-hour vesting standard, break-in-service rule, lack of reciprocity and self-pay provisions and its decisions to increase benefits.
1. Exclusion Rate
The Fund’s exclusion rate serves as evidence of the combined effect of its eligibility rules. The Fund argues that the district court misapplied the exclusion rate test by failing to take into account the high turnover in the hotel and restaurant industry. Although the Fund’s work force was highly transient, this fact has little bearing on the threshold determination of whether the exclusion rate is unusually high. The trustees’ fiduciary duties run to all participants regardless of their length of employment. The Fund’s statistics relating to employee transience become relevant only after the burden has shifted to the Fund to justify its eligibility standards.
The district court found that an exclusion rate above 97 percent was unusually high. This finding was not clearly erroneous. The burden properly shifted to the Fund to show the reasonableness of its eligibility rules. See Ponce III, 774 F.2d at 1403.
2. Reasonableness of the Fund’s Eligibility Rules
In evaluating the Fund’s justifications for the high exclusion rate, three of our decisions are on point: Ponce I, Ponce III and Miranda v. Audia. Both Ponce decisions and Miranda involved challenges to pre-ERISA, 15-year vesting rules that excluded 94 to 96 percent of all participants. *516In'no case did we engage in statistical line-drawing. Instead, we carefully considered the plan’s justifications in light of the facts.
In Miranda, we found no structural defect, grounding our conclusion on these facts:
(1) a high exclusion rate was due at least in part to the highly transient nature of construction laborers;
(2) the average age of employees qualifying for benefits had steadily decreased;
(3) the plan had adopted rules to relax eligibility requirements, including lowering the vesting age from 65 to 62 for those with 15 years of service, eliminating the retirement age requirement entirely for those with 25 years of service, relaxing its break-in-service rule, and establishing national reciprocity agreements with other plans;
(4) a reduction in the vesting requirement from 15 to 10 years would result in the vesting of an additional 445 employees out of a total of over 30,000 participants; this would not materially reduce the number excluded.
Miranda, 681 F.2d at 1125-26.
In Ponce III, we confronted a number of very similar facts. Distinguishing the plan in Miranda on the basis of the “redeeming steps [it] had taken to relax eligibility conditions,” we found a structural defect. Ponce III, 774 F.2d at 1403. The plan in Ponce III, however, had adopted two of the very rules (the reciprocity agreement and the lowering of the retirement age) that we found so salutary in Miranda. See Ponce v. Construction Laborers Pension Trust of S. Cal., 582 F.Supp. 1310, 1312-13 (C.D.Cal.1984) (Ponce II), aff'd, 774 F.2d 1401 (9th Cir.1985). In addition, while both cases involved a highly transient work force of construction laborers, we interpreted that fact differently in Ponce III than in Miranda. In Miranda, we suggested that the transience was a reasonable explanation for the high exclusion rate. 681 F.2d at 1125-26. In Ponce III, however, we viewed it as evidence that the plan’s 15-year vesting requirement was not accomplishing its asserted goal of retaining employees. 774 F.2d at 1405.
The divergent results and sometimes divergent reasoning of these decisions suggests that our inquiry must be highly sensitive to the facts and circumstances confronting the Fund’s trustees.
Before examining the justifications offered by the Fund, we turn to a statement in Ponce I that is hotly disputed by the parties and critical to our interpretation of the LMRA: “[T]he primary purpose of section 302(c)(5) is to insure that pension benefits are awarded to as many employees as is economically possible.” Ponce I, 628 F.2d at 544; see Elser, 684 F.2d at 656.
The plaintiffs interpret this statement as meaning that nothing short of economic necessity can justify excluding large numbers of plan participants. Because the Fund does not attempt to justify its high exclusion rate on financial or actuarial grounds, plaintiffs argue that “the numbers simply end the discussion.”
The Fund and amici NCCMP and the Alaska Teamster-Employer Pension Trust argue that the Ponce I statement must be read narrowly. If the statement means that providing benefits to the greatest number of participants is the compass guiding the court, they argue, then it conflicts with the underlying assumptions and goals of a defined benefit plan.
The Ponce I statement conceals a fundamental tension that affects the administration of any pension plan. Under one view, contributions to a pension plan are seen as deferred compensation, money that comes directly from employees’ pockets. The employee’s “right” to these deferred benefits is the same as his or her “right” to current wages. In a plan operating under a “pure” deferred compensation theory, employees would enjoy full and immediate vesting from the first day of employment. See Comment, Judicial Review of Fiduciary Claim Denials Under ERISA: An Alternative to the Arbitrary and Capricious Test, 71 Cornell L.Rev. 986, 1003-07 (1986).
Several courts have endorsed variants of the deferred compensation theory, pointing to statements in the legislative history of *517both the LMRA and ERISA. See Demisay v. Local 144 Nursing Home Pension Fund, 935 F.2d 528, 533 (2d Cir.1991) (employer contributions to a § 302(c)(5) plan “are in exchange for employees’ labor, just as wages are”); Knauss v. Gorman, 583 F.2d 82, 89 (3d Cir.1978) (concluding that drafters of § 302(c)(5) viewed contributions as deferred compensation); Mosley v. National Maritime Union Pension & Welfare Plan, 438 F.Supp. 413, 424 (E.D.N.Y.1977) (citing ERISA legislative history supporting deferred compensation theory).
Others see a pension plan as an incentive system that enables employers to retain and reward long-term employees. Several courts, including this one, have recognized this rationale. See Ponce I, 628 F.2d at 542 (break-in-service rule is “intended to promote an employer’s legitimate interest in the continuous employment of his employees”); Mestas v. Huge, 585 F.2d 450, 453 (10th Cir.1978); Roark v. Boyle, 439 F.2d 497, 504 (D.C.Cir.1970). ERISA’s legislative history contains some support for the incentive/reward rationale. See 1 Legislative History of ERISA, supra note 5, at 1796-97. But if a “pure” incentive/reward system were implemented, trustees would have unfettered discretion to devise standards facilitating employee retention. This they do not have.
A modern defined benefit pension plan pools contributions for all workers, using the income generated by their combined financial clout, to provide reasonable pensions for workers who satisfy reasonable eligibility standards. The formula necessarily assumes that the pensions of a significant number of employees may never vest. See Central States, S.E. & S.W. Areas Pension Fund v. Gerber Truck Serv., Inc., 870 F.2d 1148, 1154-55 (7th Cir.1989) (en banc).
A defined benefit plan is neither pure deferred compensation nor pure incentive and reward. It draws on both concepts. To apply properly the statement in Ponce I, we must recognize and balance these competing views. If our admonition that a plan award benefits to “as many employees as is economically possible” were taken literally, the deferred compensation view would triumph. Eligibility standards would be set at the lowest possible threshold and benefits would be minuscule, diminishing rewards and incentives to the point of insignificance.
We proceed cautiously then to determine whether the plan suffered from a structural defect. The Fund begins by offering four explanations for its high exclusion rate: (1) its eligibility standards were the result of discretionary policy decisions designed to retain and reward workers by reallocating contributions from short-term workers to long-term workers; (2) the minimal effect that the district court’s remedy would have on the exclusion rate demonstrates that the rate was attributable to high turnover in the hotel and restaurant industry and not to a structural defect; (3) it complied with ERISA vesting standards and implemented eligibility standards that, when adopted, were similar to or more liberal than comparable plans; and (4) it took steps to liberalize its vesting rules.
The first justification essentially relies on the incentive/reward rationale. Although this rationale may be reasonable in the abstract, our decision in Ponce III indicates that a plan must provide some evidence that vesting standards have operated to retain “career” workers. See Ponce III, 774 F.2d at 1405. The evidence before us does not show that the 10-year/15,000-hour rule significantly improved employee retention.
Although the Fund’s high turnover rate weakens its argument that the 10-year/15,000-hour requirement was rationally related to its goal of retaining career employees, the high rate gives credence to its argument that even if it had adopted five-year vesting, the exclusion rate would have been lowered only slightly. The Fund’s evidence showed that more than 50 percent of its participants left the plan with less than one year of service. Eighty-six percent of participants departed with two years or less. Employee transience was a permanent feature of the Fund; even be*518fore pipeline construction began, the exclusion rate exceeded 95 percent.
The district court’s remedy of ordering five-year vesting retroactive to 1973 reduces the Fund’s exclusion rate by a mere two percent, from 97 percent to 95 percent. If every person in the four-year category chose the self-pay option, another two percent would vest, reducing the rate to 93 percent.6 The district court found that a rate of 85 to 90 percent was typical for most multiemployer plans.
In Miranda, the 10-year standard proposed by the plaintiffs would have reduced the rate from 96 percent to approximately 94.5 percent. We endorsed the district court’s finding that this would not “materially reduce” the rate and, therefore, did not warrant finding a structural defect. Miranda, 681 F.2d at 1126; cf. Ponce II, 582 F.Supp. at 1314 (adopting a ten-year standard rather than a 15-year standard “would have significantly increased the number of participants who would have vested”) (emphasis added). We can scarcely conclude that the Fund is structurally deficient when the district court’s remedy continues to exclude over 93 percent of plan participants.
The cautionary words of Judge Anderson, concurring in Ponce I, are apt here: “The trustees are dealing with a unique laboring force and are entitled to consider the many factors which render it unique when making their judgments in carrying out the trust.” 628 F.2d at 545.
Employee turnover was unusually high in the Alaska hotel and restaurant industry. Yet, the district court discounted the transient nature of the work force, concluding that “those factors are present to some extent in every industry.” In light of the minimal impact that court’s remedy has on the exclusion rate, it would be clearly erroneous to conclude that the overriding cause of the 97 percent exclusion rate was the Fund’s restrictive vesting standards. By itself, this undermines the district court’s finding of a structural defect.
Other factors also convince us that the plan was not structurally defective. The Fund argues that its compliance with ERISA vesting standards militates in favor of finding no structural defect. Although ERISA minimum vesting standards do not displace § 302(c)(5), we must understand the nature of the legislative compromise that produced the 10-year standard. Implicit in that compromise is the notion that trustees have wide discretion to adopt or not adopt eligibility standards that are more liberal than the ERISA mínimums.
While such discretion is not limitless, a court may not intrude into the province of trustee decision-making unless a strong connection exists between arbitrary and capricious eligibility standards and the exclusion of unusually high numbers of employees. A comparison of the Ponce and Miranda decisions demonstrates that a fine line separates permissible discretionary action from arbitrary and capricious conduct. Where there is substantial doubt whether that line has been crossed, we are reminded of our institutional limitations. Altering a pension plan’s long-range actuarial assumptions can be risky business with repercussions extending to participants and times far beyond the case at hand. Judicial intervention in this complex area should be reserved for violations of the arbitrary and capricious standard which clearly cause excessive exclusion.
This is a close case. The Fund received substantial contributions on behalf of the non-vested plaintiff class, yet it repeatedly increased benefits for vested participants. Perhaps the Fund could have operated more equitably had it adopted a graded vesting schedule where participants with less than 10 years of service obtained a partial benefit. Alternatively, a savings plan might have distributed funds to more participants.
We cannot say, however, that the Fund’s failure to amend its standards so that another two to four percent of its participants *519could vest warrants judicial intervention. During the years immediately after passage of ERISA, few plans, if any, provided for more liberal provisions than the minimum standards set forth in ERISA. See Rehon, The Pension Expectation as Constitutional Property, 8 Hastings Const. L.Q. 158, 195 (1980). Moreover, the Fund’s actuary testified that he cautioned against expanding benefit coverage during the pipeline construction era. Trustee conduct that is consistent with reliance on the advice of an expert actuary supports a finding of no violation under § 302(c)(5). Souza v. Trustees of W. Conference of Teamsters Pension Trust, 663 F.2d 942, 947 (9th Cir.1981).
In finding arbitrary and capricious conduct, the district court relied partially on its finding that a “significant number” of the Fund’s vested retirees had income replacement ratios exceeding 100 percent. This finding appears to be based on evidence introduced by the Fund of projected pensions for two hypothetical retirees. What the court meant by a “significant number” is not entirely clear,7 but the evidence of excessive income-replacement ratios does not appear to be as strong here as it was in Ponce II. In that case, the district court found that benefit levels of vested retirees “far exceeded recognized standards” and that the ratio was “usually” in excess of 100 percent.8 Ponce II, 582 F.Supp. at 1316, 1318.
The district court also relied on its finding that the Fund’s break-in-service rule contributed to the high exclusion rate by causing involuntary forfeitures of hours when employees were unable to find work that was covered under the Fund. There is no showing, however, that elimination or modification of the break-in-service rule would so substantially reduce the exclusion rate that it was arbitrary of the Fund to retain the rule.9
No additional role played by the rule in discouraging some workers from pursuing covered employment appears to have been critical. We think the district court’s reliance on the effects of the rule was secondary to its real concern — the operation of the 10-year/15,000-hour rule.
The trustees’ failure to adopt reciprocity or self-pay provisions was not arbitrary and capricious. As the district court noted, reciprocity agreements did not exist in the hotel and restaurant industry. As for a self-pay provision, the plaintiffs’ own expert could identify only two pension plans that had this option.
We conclude that the trustees acted within their discretion in maintaining the Fund’s eligibility standards, perhaps near the limits of that discretion. The district court did not consider adequately the Fund’s evidence that the highly transient work patterns of employees in the hotel and restaurant industry skewed the exclusion rate. Judicial intervention in pension plan decision-making is inappropriate without a stronger nexus between trustee conduct and the exclusion of unusually high numbers of participants.10
*520III. Statute of Limitation
Although the foregoing disposes of the appeal on the merits, to avoid any inference that we approve of the district court’s disposition of the statute of limitation issue, we address it briefly.
This case is controlled by the ERISA statute of limitation, 29 U.S.C. § 1113.11 See Kwak v. Joyce, 683 F.Supp. 1546, 1549 (N.D.Ill.1988). The Fund argued that the named plaintiffs did not satisfy the statute because each had “actual knowledge” of the trustees’ alleged breach of fiduciary duty more than three years before this suit was filed. In rejecting the Fund’s argument, the district court characterized the trustees’ failure to relax eligibility rules as “a continuing breach of fiduciary duty and statutory violation.” It concluded:
If past unpaid benefits were sought in this case, the period of their availability, if any, would be limited by the ERISA statute of limitations. However, only prospective relief is sought — i.e., the vesting now of the pension rights of those unlawfully excluded so that they will have benefits from this point forward. That being so, and the violation having continued to and after the date the suit was filed, the statute of limitations does not bar any part of the claim.
The court’s reasoning is based on a variation of the continuing violation theory employed most often in employment discrimination and antitrust cases. See, e.g., Sosa v. Hiraoka, 920 F.2d 1451, 1455-56 (9th Cir.1990) (employment discrimination); Eichman v. Fotomat Corp., 871 F.2d 784, 793-94 (9th Cir.) (antitrust), amended on other grounds and superseded, 880 F.2d 149 (9th Cir.1989). Its application of that theory hinged on this concept of the alleged breach: the trustees were faced with numerous decisions where they could have relaxed vesting rules; they chose instead to maintain the status quo or to increase benefits for vested participants. The result was a series of discrete but related breaches. As each new breach occurred, the three-year limitation period of § 1113(2) began anew. Because the alleged breaches continued up to and after the filing of the suit, the three-year period was no bar.
This application of the continuing violation theory founders on the plain language of § 1113(a)(2). This section requires the plaintiff’s knowledge to be measured from the “earliest date” on which he or she knew of the breach. While the trustees’ conduct may be viewed as a series of breaches, all were of the same character: a failure to amend vesting rules. Once a plaintiff knew of one breach, an awareness of later breaches would impart nothing materially new. Cf Pace Indus., Inc. v. Three Phoenix Co., 813 F.2d 234, 237-38 (9th Cir.1987) (in antitrust cases, the continuing violation theory requires an overt act that is new and independent, not merely a reaffirmation of a previous act; it must inflict new and accumulating injury on plaintiff). The earliest date on which a plaintiff became aware of any breach would thus start the limitation period of § 1113(a)(2) running. The district court’s application of the continuing violation theory essentially reads the “actual knowledge” standard out of the statute.
Our analysis of § 1113 in Ziegler v. Connecticut Gen. Life Ins. Co., 916 F.2d 548 (9th Cir.1990), is not to the contrary. In Ziegler, we analyzed separately the issues of when a cause of action accrues under § 1113 and when the limitation period begins to run. See id. at 550. The latter determination requires the court to apply *521the three- or six-year standard in the statute. Id. at 552. A continuous series of breaches may allow a plaintiff to argue that a new cause of action accrues with each new breach. But if the breaches are of the same kind and nature and the plaintiff had actual knowledge of one of them more than three years before commencing suit, § 1113(a)(2) bars the action.
The record here indicates that, as of the late 1970s and early 1980s, at least one, and perhaps all, of the named plaintiffs may have had actual knowledge that the trustees had failed to relax vesting provisions so that more workers would benefit. The district court’s erroneous application of the continuing violation theory prevented it from deciding this factual question. Our disposition of the merits makes its resolution unnecessary.
IV. Attorneys’ Fees
The district court awarded attorneys’ fees to the plaintiffs under the ERISA fee-shifting provision, 29 U.S.C. § 1132(g)(1), which provides that a “court in its discretion may allow a reasonable attorney’s fee and costs of action to either party.” We have said in dictum that this language permits an award of fees to a non-prevailing party. See D’Emanuele v. Montgomery Ward & Co., Inc., 904 F.2d 1379, 1382-83 (9th Cir.1990). But see Bittner v. Sadoff & Rudoy Indus., 728 F.2d 820, 829 (7th Cir.1984) (losing party may not obtain attorney’s fees under § 1132(g)(1)); Fase v. Seafarers Welfare & Pension Plan, 589 F.2d 112, 116 (2nd Cir.1978) (“An altogether sufficient support for the court’s decision not to award attorney’s fees under ERISA is that the attorney obtained no relief under the statute.”); see also Ruckelshaus v. Sierra Club, 463 U.S. 680, 103 S.Ct. 3274, 77 L.Ed.2d 938 (1983) (holding that analogous language in attorney’s fee provision of Clean Air Act does not permit court to award fees to wholly unsuccessful litigant).
Even if § 1132(g)(1) permits an award of fees to a losing party, such an award would not be justified here. Under Hummel v. S.E. Rykoff & Co., 634 F.2d 446, 453 (9th Cir.1980), we consider five factors in determining whether to award fees under § 1132(g)(1):
(1) the culpability and good faith of the opposing party;
(2) the ability of the party to pay fees;
(3) the increased deterrence that would result from the award;
(4) whether a number of beneficiaries would benefit from the award of fees; and
(5) the relative merits of the parties’ positions.
We conclude that all but the fourth factor favor a decision not to award fees. The fee award by the district court is reversed.
CONCLUSION
In light of the foregoing, we need not reach the other issues raised by the plaintiff class in its cross-appeal. The trustees’ maintenance of restrictive vesting standards was not arbitrary and capricious and the Fund was not structurally defective. Plaintiffs are not entitled to attorneys’ fees.
REVERSED.
. The three named plaintiffs had accumulated these hours of credited service: Joe Ahlers, 13,-914 hours; Helen Phillips; 6,020 hours; Carson Workman, 13,542 hours.
. Some pension plans have reciprocity agreements with other plans that permit workers who move from one geographic area to another to avoid forfeiture of years of credited service. The court acknowledged that no such agreements exist in the hotel and restaurant industry.
.Some pension plans have self-pay provisions that permit participants to avoid forfeiture under a break-in-service rule by making their own contributions when they are without covered employment.
. The ten-year minimum applies to multiem-ployer plans with collective bargaining agreements. Other types of plans must comply with shorter mínimums. See 29 U.S.C. § 1053(a)(2)(A), (B), (C) (1988).
. During the floor debate of a precursor bill to ERISA, Senator Bentsen stated:
We have to remember that these are voluntary pension plans ... and one of the motivating reasons for putting in the plans is that employers want the employees to stay with them for a while, because they are going to spend money training those employees.... Then, we have to look at the fact that if we require very early vesting, that means that the employer has to allocate more of his pension contributions to younger employees who have a high turnover rate. As a result, the older employees will be penalized.
Subcomm. on Labor of the Sen. Comm, on Labor and Public Welfare, 94th Cong., 2d Sess., 1 Legislative History of the Employee Retirement Income Security Act of 1974, at 1796-97 (1976) [hereinafter Legislative History of ERISA ].
. The district court calculated that five-year vesting would award benefits to up to 1,002 members of the plaintiff class. The Fund calculates that another 1,070 might be eligible for the self-pay option.
. A list of the Fund's pensioners shows that the majority receive less than $1,000 per month. It does not provide their income-replacement ratios.
. The Ponce II court found income-replacement ratios that averaged between 91 and 110 percent. 582 F.Supp. at 1315. The trustees in that case had expressly rejected a ten-year standard and corresponding monthly pension benefit calculated by the plan’s actuary. Id. at 1313. The court found that this monthly benefit would have resulted in an income replacement ratio that averaged between 69 and 80 percent. Id. at 1315. These findings are more detailed and indicative of arbitrary and capricious conduct than the findings of the district court here.
. The attack on the break-in-service rule in this case is based entirely on its alleged effect in excluding high percentages of workers from benefits. No argument has been made that the break-in-service rule is arbitrary simply because it makes no allowance for involuntariness of the break. See Bolton v. Construction Laborers’ Pension Trust, No. 90-55654, slip op. 12211 (9th Cir. Aug. 30, 1991).
. The district court’s conclusions of law focused solely on the LMRA. The plaintiffs also assert that the Fund breached its fiduciary duties under ERISA. We reach the same result under ERISA § 404. The same standards apply to actions under both § 404 of ERISA and § 302(c)(5) of the LMRA. See Music v. Western *520Conference of Teamsters Pension Trust Fund, 712 F.2d 413, 418 (9th Cir.1983).
. Section 1113 provides:
No action may be commenced ... with respect to a fiduciary's breach of any responsibility, duty, or obligation ... after the earlier of—
(1) six years after (A) the date of the last action which constituted a part of the breach of the violation, or (B) in the case of an omission, the latest date on which the fiduciary could have cured the breach or violation, or
(2) three years after the earliest date on which the plaintiff had actual knowledge of the breach or violation....
29 U.S.C. § 1113 (1988), amended by Pub.L. No. 101-239, title VII, §§ 7881(j)(4), 7894(e)(5), 103 stat. 2443, 2450 (1989).