[corysama]

Similarly, my general rule for optimization is: Add up all the time that will ever be spent waiting on this code by all users ever. If that's less than the time required to implement the optimization, don't bother.

That eliminates a surprising number of tempting gold-plating situations that occur in rarely-executed code.

On the other hand, cutting 1 second of wait from a million daily active users adds up to whole human lifetimes saved rather quickly.