I know this is not the best question, I'm just trying to understand what needs improvement.
I just read a Phoronix article, which compared the FOSS radeon drivers a 5 years old FGLRX catalyst. As you would expect FGLRX was multiple times faster, even the feature set was not completely implemented.
The big question, not answered in the article, was why? I noticed FGLRX brings its own libGL, does Nvidia do this also? I know hardware registers are not always completely known, and yadda yadda... I still suspect that mesa is not a strong performer.
What needs to be done to reach remotely close to catalyst speed? What projects need help? Which ones need to be completely ditched?