This is why Seth Godin champions NOBS instead of a traditional business school:
http://sethgodin.typepad.com/seths_blog/2007/04/nobs_the_end...
NOBS maximizes #1 and #2 with little work. Now if only there really was a NOBS. Wait, it's called YCombinator! :)
In this context, we can say there are three things that can determine someone's success:
1) confidence others have in you
2) self-confidence
3) actual ability
Going to a "good" college affects 1 and 2 the most, and arguably 3 the least. In a startup 3 and 2 are arguably the most important, and 1 the least - so I guess that agrees with pg's argument. In the corporate world, 1 and 2 matter the most, and 3 the least - aligned with what a brand-name college provides.
So the argument seems a bit circular: based on the yc sample set, college doesn't matter because college doesn't matter to the yc sample set. Even if that SHOULD extend to the world at large, it doesn't.