How is it not? Seems like simple economics to me. Something like maintaining a custom kernel distribution takes ${x} amount of engineer time for ${y} amount of “savings” (better perf, custom patches, nonstandard features, whatever). Why bother until you have a fleet of machines large enough for y to be greater than x?

I don't have numbers but I feel this is exaggerated. Most people work on pretty normal stuff.

The biggest proprietary sink for most people is stuff like the build system/tooling/etc. Not many people have to be in a role where you're micro-optimizing something. They certainly exist, but not in the numbers some people on here seem to think.