Neat! I wonder whether analogous differences occur in speech (one would expect so); if that's the case it would provide one way to improve on existing speech recognition technology (which relies heavily on bigrams/trigrams/n-grams to figure out what the next word is likely to be).
I wouldn't call an 80% success rate for predicting author gender 'ferocious predictive power', though. 80% is good, but not astounding.

This is the part that got me:

These differences are significant enough that even when Koppel's team analyzed scientific papers-which would seem to be as content-neutral as you can get-they could still spot male and female authors.