Stop word distributions are also sufficient for identifying gender in most English corpora with about 80% accuracy -- and you don't need much text, either.

[Edit to add linky: http://u.cs.biu.ac.il/~koppel/papers/male-female-text-final.... ]