Linus's point is that when he documents a feature "unstable" or "do not use", distros often use it anyway. Once that happens, it doesn't matter how it's documented, the interface must be preserved forever.
It's not "Linus finds warnings in documentation useless" as you suggest, it's "Linux users have demonstrated time and time again that it's useless". A subtle but important difference.
The only practical solution so far has been to simply withhold the feature until the interface is stable. But all programmers do this. I don't see why you find that so deeply disturbing...?