Yes, I know performance can be critical for some products - and that's the case for Stack Overflow. But I think you can't simply ignore other aspects like maintainability or security.
This article from Sam Saffron is very interesting. He was very reasonable testing the ORMs available, and only after realizing that none reached the performance he needed, he decided to write his own. The problem is when people do that and "reinvent the wheel" without searching if someone already solved that problem for you.