Is it possible that what we're seeing here is a reversion to the mean[1] - CEOs who do particularly well some years get large pay packages, and then when their performance reverts to the mean in subsequent years they end up overpaid relative to performance?
And Tournament Theory: compensation is often based not on proportionate productivity, but rank order, which may be easier to measure, and serves as probabilistic-motivation for other workers competing for the top spot.
You beat me to it. Another way to state it would be that having a richly-paid CEO is a lagging indicator; or the companies with the resources to have a trophy CEO are likely already topped out and running up against the law of large numbers. Imagine being Tim Cook, CEO of Apple: he ought to expect high pay for such a prestigious job (I haven't looked it up), but continuing to grow the stock price at past rates would be next to impossible.
I'm sure they ruled out the more obvious confounding variables. All else being equal, you would expect a result like they got, based on the null hypothesis that expensive CEOs are at least average among CEOs, combined with the suspicion that their hiring is not a random event. Factor in that the researchers were undoubtedly biased in favor of the outcome they found, and a lot of skepticism is justified in making sense of this.
[1] Thinking, Fast and Slow by Daniel Kahneman