On optimizing the signal-to-noise ratio of two-state experiments