What about Salman Khan's TED talk, where he shows that it is a frequent occurrence that kids stumble on a math concept, but then rapidly catch up or exceed their peers. If learning as a general phenomenon occurs in plateaus, but turns out linearly the same in the long run for most everybody (like a graph of the prime counting function seems choppy up close but a line out far), perhaps it is the case we intuitively think some people have a talent, because they didn't experience any early plateaus, but experience some later instead, when they aren't under as careful scrutiny. In other words, it might be a statistical phenomenon of the learning progression (assuming the plateaus are uniformly randomly distributed). Having thought a lot about this issue, I think it is closer to the truth.

In any case, I second the request for more studies.