I recently read Malcolm Gladwell's Goliath book (which I don't particularly recommend) which discussed some of this research. He suggested that class size was an example of a phenomenon he was interested in, that many good things (like reducing class size, if you start at 100) become bad eventually (eg it's hard to get a range of opinions or friends in a class of 10). He claimed that some of the positive research is contaminated by supportive parents managing to get their children into small classes. He didn't, to my mind, go nearly far enough into how hard it is to do good research on this topic. For one thing, the good practice of randomisation is in tension with the obviously important factor of how prepared the teacher is to adapt the teaching style.
