How does this deal with studies (I confess I don't have links, I read about them in Bad Science by Ben Goldacre, mentioned in this blog post I found) that show that two sugar pills 'work' better than one, and a saltwater injection 'works' better still at relieving symptoms, that different colours have different levels of effectiveness or even that one can induce negative side-effects (e.g. vomiting) simply by telling the subject that they exist.

None of these are at all explainable by regression to the mean, diseases running their course or increased compliance with medical instructions. I would even be surprised if there was biased self-reporting going on, unless the groups were aware of each-other's existence I doubt the one-pillers were explicitly thinking "only one pill, I probably won't get much better".

This seems to point strongly towards a direct causal link between believing you are going to get better and getting better.