Your power to detect an effect is dependent upon the effect size. For example, the power for a T-test is:

Where, phi() is the probit function, tau is the effect size, n is the number of samples and sigma hat sub D the standard deviation.

In this case the equation would be different (I'd have to look into the details of how MuTect works to get the exact equation), but the principle the same. I should note that the interrelation between effect size (allelic fraction) and power is actually intuitive. Suppose you wanted to be powered to detect very rare changes. How do you thing your coverage would need to change? Intuitively, it'd need to go up. Similarly, if you only cared about huge changes you would suspect that you wouldn't need much coverage. In this case, one's intuition and the actual equations are in agreement.