The notion that abilities de-correlate at high range is known as Spearman's law of diminishing returns . A simple analogy from sports: ability to run marathon positively correlates with the ability to sprint, in the general public, but among the world class athletes, you run into genetic variations which trade sprinting performance for marathon performance and vice versa.

Another point is that IQ tests have to be neutral with regards to the background skills or knowledge, which has a very unfortunate side effect of not measuring performance of the mechanisms involved in forming or applying skills and knowledge. By the way, on a Gaussian prior, poor correlation implies very substantial regression towards the mean.

edit: cut-n-paste error in URL.

wouldn't that make the extreme end of IQ results less and less correlated with actual intelligence?

It seems much more plausible that at extreme intelligence, the correlation to life outcomes starts to break down. Once you earn enough money to live comfortably, it probably leads to more life satisfaction to spend your time on leisure, rather than earning more, and in particular, the cleverer someone is the more we might expect them to realize that's the tradeoff.