This is a nice example for the Bayes theorem: what's the probability of seeing a computer running Windows "in the wild", given that "the wild" comprises the computers of a startup crowd located in the same area. When you're limiting the universe to such a small fraction of similar samples, you're bound to get skewed results.
I think most college students will agree that they see windows every single day
Certainly PG isn't the only one who doesn't see very many Windows computers, and I think this has probably been increasingly true over the last ~9 years since this piece was written.