The conclusion to draw, then, should be a more general one: given an observation O and an explanation E of O, people can over-weight E as a piece of evidence about O's probability.

But is it overweighting to use the fact that the explanation is bad as evidence against the statement being true? A true statement is more likely to have a good explanation than a false one, so it seems that one could do a Bayseian update on the truth of the staement based on the quality of the explanation.