Developer time is a zero sum game. If one of these optimizations bites me and I spend one day figuring out what happened, I don't spend one day doing optimization somewhere else, where a much greater speedup can be achieved. Or, to out it in your money difference terms, I could be adding value to the program somewhere else, adding or polishing features.

It's a balancing game. Looking at it through the lens of % speedup, ignoring everything else is just stupid. In a perfect world we'd make decisions based on numbers. He provided his. Where are yours?