Comment by Zigurd

The description sounds accurate compared to what I heard when I was in contact with Microsoft insiders. That was about 15 years ago. What's surprising to me is that it is still this way.

A lot of the justification for being exceedingly conservative about performance improvements was based on the difficulty of doing testing that is sufficient to leave no doubt that changes don't cause regressions nor any side effects bearing on compatibility.

I'm surprised that hasn't changed, or that the value equation about possibly breaking customer software hasn't changed. By now this approach must be a very entrenched culture.