However its a valid optimisation and might be worth tweaking in.
I'd like to stress that dampening is impossible
if the 'sum' is an aggregrated value (used for calculating 'avg'). It will leak time, and can't recover. Look very carefully at the code, the dampening is applied to the values in the slots, which means that the dampening will (slowly) disappear after 10 frames. This handles both erratic delays and poor resolution timers. Please don't optimize for optimizations-sake. The current code is correct and extremely fast. The 'correct' part is most important.
It can be a few nanoseconds faster, indeed, but at the expense of reliability and correctness! There are enough screwed up sync( ) implementations, I'd prefer to keep this implementation correct.
Don't get me wrong, dr_evil: if you think your implementation is better, taking into account a vast number of corner cases, I'd be very interested.