Sobering (but thus useful) experimental findings on bias in teaching evaluations: www.tandfonline.com/doi/abs/10.1...

would be interested in variation by academic discipline. I'd wager that a few specific disciplines tend to skew the overall picture.