My recent post on 'Desirable vs. Rationality-Enhancing Dispositions' contains an important mistake. I began by distinguishing 'rationality-enhancing' dispositions (desirable in virtue of conducing to good actions) from dispositions that have other good effects. I then suggested, mistakenly, that rational status is always transmitted from rationality-enhancing dispositions to the particular actions they dispose one to perform. This transmission principle holds only when the 'rationality-enhancing' disposition serves as a generally reliable 'rule of thumb', like the rule against killing people even when murder might (prima facie) seem to promote utility. But in non-ideal agents, a disposition might also conduce to good actions for a very different kind of reason, as I will illustrate below.
Meet Cam, a callous consequentialist. Cam is one of those utilitarians who likes humanity but not people so much. Due to his lack of regard for those around him, he tends to act insensitively, and makes other people (not least his poor family) miserable. Upon reflection, Cam recognizes this to be unfortunate. So he takes a pill that makes him a much more caring and loving person. He is now disposed to attend to the welfare of those that are close to him. This causes him to act in much better ways: in particular, he finds it easier to refrain from making the kinds of insensitive remarks that previously caused so much harm. The one downside is that he is now much less inclined than before to donate to charities like Oxfam that promote the impartial good. He would rather spend that money on his family. This is bad, but (let's suppose) not nearly as bad as Cam's callous actions had been.
The structure of the case is that Cam was previously weak-willed in a very bad way. He then acquired a disposition that helped him overcome this weak will, and so perform better actions -- though at the cost of acquiring a new (less bad) weakness. Because it is clear (by stipulation) which of his newly-disposed actions are better than before and which are worse, the epistemic argument for following a generally beneficial disposition no longer applies (in the 'clearly worse' case). So, it's right for Cam to acquire this disposition in virtue of the actions it conduces to in general, but some particular actions it conduces to may still be considered wrong (or less than perfectly rational/beneficent). We can thus have cases of what Parfit would call 'blameless wrongdoing' without having to appeal to dispositions that have good effects besides action. (Of course, the disposition may also be good for other reasons -- the point is just that my case doesn't rely on this.)