Variation may be the stuff of evolution, but many adaptationist arguments predict an optimal phenotype, with modest variation driven by mutation. Fitness peaks for the optimum with modest variability around this peak. In fact, I show that one component of fitness -- the Lifetime Reproductive Success LRS -- has a dramatically skew distribution even for a single phenotype. The high variation in LRS means that random effects can play a bigger role than expected in the chance of survival of an advantageous mutant. A branching process analysis shows that this chance depends on the odds of having no offspring versus one, rather than on the mean LRS and its variance. These aspects of high variability show that (i) selection has to overcome a substantial diffusive effect, (ii) we need great care in assuming that quality differences can be based on LRS as a performance index.