> I'm afraid that the incentives are to credit models that predict more warming just a bit more than those which predict less; the expected effect would be that the predicted warming effect would get larger over time.

This actually doesn't make any sense! In your valued Milikan example, the problem was subsequent experiments were biased towards lower values, meaning that for climate models the bias would be expected towards under heating models. In other words, the lesson from Milikan is that there is a risk our models aren't heating fast enough.

So by the quoted example we should be worry that our models under predicts, and yet Milikan is stood up to argue the models are too wild in their warming predictions?