Wait this is actually brilliant in a couple of ways, because to get the right (estimated) answer, the listener has to distinguish between probability that one of the three is a rabbi and this is a joke, and probability that this is a joke if we put the probability of the third being a rabbi at 100%.

It follows the setup of a rationality calibration question while subverting it and rendering "guessing the teacher's password" useless, since c) is (maybe) higher than a) or b)