Fair enough, but my point remains that the programmers are implementing a spec designed by someone else. When your hackers are completely in control, you can judge them simply by their ability to get things done. If they're implementing someone else's spec, then measurement isn't that simple because their performance will be heavily influenced by the quality of the spec, no matter how much veto power you give them. So, judging by skill at optimizing might well be the next-best alternative.

Then again, maybe we're all being ridiculously overanalytical and Gates just meant "improving" when he said "optimizing".

'...Gates just meant "improving" when he said "optimizing".'

I think that's the case.