Many plants, insects, and crustaceans show within-population variability in dormancy length. The question of whether such variability corresponds to a genetic polymorphism of pure strategies or a mixed bet-hedging strategy, and how the level of phenotypic variability can evolve remain unknown for most species. Using an eco-genetic model rooted in a 25-year ecological field study of a Chestnut weevil, Curculio elephas, we show that its diapause-duration variability is more likely to have evolved by the spread of a bet-hedging strategy than by the establishment of a genetic polymorphism. Investigating further the adaptive dynamics of diapause-duration variability, we find two unanticipated patterns of general interest. First, there is a trade-off between the ability of bet-hedging strategies to persist on an ecological time scale and their ability to invade. The optimal strategy (in terms of persistence) cannot invade, whereas suboptimal bet-hedgers are good invaders. Second, we describe an original evolutionary dynamics where each bet-hedging strategy (defined by its rate of prolonged diapause) resists invasion by all others, so that the first type of bet-hedger to appear persists on an evolutionary time scale. Such "evolutionary priority effect" could drive the evolution of maladapted levels of diapause-duration variability.