I’m not sure he assuming some “fixed cost” so much as pointing out (or at least arguing) that such a model explains the data as well the “just so” conclusion. Additionally, he is suggesting that such a model, in which agents respond to incentives rather than just have “true” preferences is stronger.