Right, and at least I am not aware of complaints like such. My experience rather is, that any larger customisations of open source software are usually propagated back to the project for the simple reason of saving the hassle of maintaining a fork of that software.

I imagine you don't have much experience with enterprise code.

They have piles of patches and team specific forks that are never going to go out of those Clearcase and TFS repositories.

Sony was the only company I ever saw doing a presentation at a LLVM summit, referring that keeping up with upstream as their main reason to get the PS4 clang code "partially" into clang. They did not contribute back any kind of optimizations that might reveal too much of PS4's architecture.

I guess some would say it is better than nothing.

OEMs using Android or BSD/Linux forks for embedded use, don't have any issue maintaining their own forks.

I am working at a very large software company, don't know whether that qualifies as enterprise code :). Yes, if you do customisations of open source software which could reveal trade secrets, I wouldn't expect this to be back-donated, and you could do that only with permissive licenses in the first place. But the large majority of open source software means just using it, and if it gets modified at all, it consists of bug fixes an small enhancements, that can and are back-donated to the projects. And as you wrote yourself, Sony tries to get the changes which don't disclose their trade secrets into upstream to reduce the merging load.

How many of those patches have non zero chance to be accepted and how much time would it take to actually hop through all obstacles to push them? My answers are 1.) a few 2.) a lot.

Majority of Linux contributions are paid by various companies. It is not volunteers working for free, it is companies paying.