The issue, for many companies, is that the perks the high performers want in exchange for this "arbitrage" that exists in hiring them-- very high levels of autonomy, rapid career advancement, implicit trust from the first day of work-- are considered unaffordable in the actual political context of the company.

Companies won't give high performers 10x salaries because the high performer might not deliver if the conditions are bad, but if they give them high degrees of autonomy and interesting projects, that's visible to the less adept workers and it becomes an issue for management. Managers know that if they let their best people follow different rules, everyone will expect the same freedom.

High and low performance don't seem to be intrinsic to people, in the sense that mediocre people can turn excellent given the right conditions, and vice versa. What does seem to be intrinsic is that there are high- and low-variance people. Low variance people are reliable ladder-climbers, and high variance people are the creative ones who might have a mediocre year or two, but then have a huge breakthrough. The top performers are, for the most part, high variance individuals. High-variance people can also fail quite badly, in certain circumstances, so companies are very nervous about hiring them at all. The "corporate ladder" rewards reliable mediocrity, not intermittent excellence.

Intimidation-based management (which has been dominant since 1800) regresses performance to the mean, and its purpose is to reduce variance. When the input-output curves relating skill and effort to productivity were concave, variance-reduction was the way to go, because lower variance was equivalent to higher performance. Now that we're in a convex world (at least in software) variance-reduction fails us.