Your statistics are only valid if the patch rate for each stable series is constant over time. It seems more likely that patches flow in the fastest at the beginning of the stable series, and then the rate drops off over time. If that's the case, then your calculation will always show the most recently released kernel as being the worst, because it's the one that's still in that early part of the cycle. Better would be to compare the same period (the first 61 days) of each kernel.