Abstract: Dormancy is a widespread adaptive strategy that enables biological populations to persist in fluctuating environments. Yet how its evolutionary benefits depend on the temporal structure of environmental variability, and whether dormancy can become systematically maladaptive, remains poorly understood. Here we examine how dormancy interacts with environmental correlation times using a parsimonious delayed-logistic model in which dormant individuals reactivate after a fixed lag while birth rates fluctuate under temporally correlated stochasticity. Numerical simulations and analytical calculations reveal that the joint effect of demographic memory and colored multiplicative noise generates a strongly non-monotonic dependence of fitness on dormancy duration, with three distinct regimes of population performance. Very short dormancy maximizes linear growth but amplifies fluctuations and extinction risk. Very long dormancy buffers environmental variability, substantially increasing mean extinction times despite slower growth. Strikingly, and central to our results, there is a broad band of intermediate dormancy durations that is maladaptive, simultaneously reducing both growth and persistence—an effect that arises generically from the mismatch between delay times and environmental autocorrelation. The predicted bistability between short- and long-dormancy strategies is confirmed in an evolutionary agent-based model, which avoids intermediate lag times and selects for evolutionarily stable extremes. Our results show that dormancy duration is not merely a life-history parameter but an adaptive mechanism tuned to environmental timescales, and that “dangerous middle” dormancy times can be inherently disfavored, with implications for understanding persistence in seed banks, microbial persisters, and cancer cell dormancy. More broadly, this work identifies a general mechanism by which demographic delays interacting with correlated environmental variability generate a non-monotonic fitness landscape that selects for extreme timing strategies, and raises fundamental questions on analyzing delayed, non-Markovian dynamics driven by correlated multiplicative noise near absorbing boundaries.