Performance is all about moderation and deliberate decisions, and Paul Vick nails it again. People get too obsessed about the "perfectly" performing systems. All too often there is a "good enough" metric for performance. Trying to make a window appear in 50ms instead of 150ms is probably not worth it, given that the delta is almost imperceptable to most people... of course, if this turns into a problem where your complex window takes 1500ms then you are in trouble.

Performance must be combined with product scenarios to have any meat.