Allocating Variance to Maximize Expectation