Imagine that you are responsible for a department of people, all of whom need to work together to accomplish a big job. Assume that talent is normally distributed - maybe you have 1 superstar and 1 idiot, 10 high performers and 10 laggards, and 100 average developers. Also assume that the boundaries between categories are fuzzy - it's not that there's 1 superstar who absolutely stands out, it's that there are 2-3 people who are really good, and one is obviously better to you but that isn't apparent to everyone else, and another 20 or so people are just a bit behind them.
What do you do?
If you preference your best programmer, give him the pick of assignments and a lot of latitude to choose projects, you will demotivate all the people who are just below him. Ironically, a lot of michaelochurch's past comments are examples of this - his perspective on the "Real Googler line", and even the grandparent comment, is what things look like for someone who is talented and high-performing and yet right below the line. These folks become bitter, demotivated, subversive, and ultimately toxic to the organization.
Moreover, you destroy the effectiveness of your superstars as well. To accomplish anything in a large organization, you need cooperation from your peers. When your peers resent your special treatment or want to sabotage your rise, you won't get this. So the only way to succeed as a superstar in a large organization is to act very subtly and bring a lot of people with you, which usually means keeping your more radical ideas to yourself and only saying things that can be heard. (Which, really, is how it should be - think of how unpleasant it is to work in an environment with a "golden kid" who's always running around with the latest technologies, not answerable to anyone else, and yet has management's favor.)
This is also why getting rid of laggards is far more important for organizational health than hiring superstars. If you populate an organization full of folks who understand compilers, or understand immutability, or can work with multiple languages, or who understand design patterns, then those techniques become part of the general culture that the whole eng organization can use. If you lack that critical mass, any superstars that understand why those concepts are important will not have the freedom to use those techniques because of the political issues above.
There's very good evidence that the middle-managers know exactly who their star performers are. When they quit to found their own company, who do they call? You often have groups of programmers that have worked together through multiple companies, and as soon as you hire one the whole group comes on board within a year. But within a company, the constraints of organizational politics prevent them from favoritism, which is probably how it should be.