Well, I used to program including some (very rudimentary) OpenGL. Does that count?
One thing I did note was that in order to make Cg look even better they compared it to assembly. The problem with any high-level language is that it will never be as fast as hardcoded assembly (short of optimization). Of course, it's much easier to program with a high-level language.
There are a lot of neato assembly tricks, even if I forgot a lot of them now.