The context: I am currently in Princeton at a Park City Mathematics Institute (NSF-supported, high-school teachers through research probabilists--really neat. Jim King has been a co-organizer of the whole series.) Earlier in the week we had a joint banquet with the African-American Research Mathematicians Conference. I had the good fortune to be seated at a table with Scott Williams, a topologist from SUNY at Buffalo who told the following tale:
This year he had two sections of business calculus to teach. At the beginning of the semester he set himself a policy which he maintained consistently. Section B he taught flawlessly, with the material flowing smoothly onto the board and every i dotted and t crossed. For section A he prepared equally carefully, but planted errors and breakdowns--several per day. If he got "stuck" he would hold out for the class to bail him out, and if they hadn't by the time the bell rang he would point out that the material would be on the exam and they had better help him figure it out the next day. Student evaluations for section A were every bit as scathing as one would expect them to be. Their mathematical performance, on the other hand, was consistently better--on identical tests given earlier in the morning, which doesn't leave much room for ambiguity.