I've found that I inherently think that I've screwed up majorly in my code. Evidence shows though that's not the case, but that doesn't change how I feel. As my own strongest critic I write some pretty strong test cases, because I'm sure I did something wrong. Of course I can't test my own test plan, so I'm always worried that I missed something stupid where I should have known better.
I also don't like strong emotional confrontation.
Can you see why I wouldn't want to work with someone abrasive like Torvalds? Part of me will always be on edge expecting to be chastised, even if that were never directed to me. So it isn't only the targets of his abrasion which are affected, but also those who think they might be targets, even if that belief is wrongly held.
I genuinely have no special insight into what Linus is like as a human being.
I just think that his abrasiveness (even if it only manifests itself in a few circumstances) is a negative that people tolerate because he is such an exceptional project manager and programmer. The keyword is 'tolerate'.
Too often, people think that his success comes not from his technical mastery (which is very difficult to imitate) but from aggressive rants and call outs - and try to imitate those and justify it as 'well that's how Linus runs the kernel'. It's even more prevalent with people cargo culting Steve Jobs' personality flaws thinking it will make their company the next apple.
People don't just tolerate him for being negative, because he's really not all that negative. He's helpful if people actually need help. However, if he trusts you to not be stupid, and you break his trust, he's going to get mad.
He needs to be able to trust people because he needs to merge thousands of patches and make sure nothing bad gets in.
This is the only reason you should ever do what Jobs or Torvalds have done: trust.
We haven't seen Carmack when someone tries to contribute naff code to one of his game engines. He doesn't develop in the open the way Linus does.