I think the problem with optimization as an approach to getting a program to be fast is that it's fairly often not bottlenecks that are the problem, but architectural choices that don't show up in the profiler's flame charts.

Even if you find a slow function that constitutes 20% of the runtime, and optimize the living hell out of it to cut out 20% of the execution time, guess what your program is now about 4.1% faster.