What empirical evidence is that? Most people don't understand the idea of the regression to the mean (check out the Kahneman quote here at a href="https://en.wikipedia.org/wiki/Regression_toward_the_mean#Regression_fallacies">here if you want a good example of why punishment seems to improve performance, but doesn't really). Also, as a long-time Linux user (since 1994) I have to say that as much as I like the OS, its success has been largely based on the fact that it was mostly the only game in town in terms of free UNIX systems at the time it came out, and was able to capitalize on that momentum since. The free BSD variants were having legal issues that a lot of people thought would sink them, Minix wasn't really a full OS, and everything else that would run on x86 software (and there weren't many) cost money. But at the end of the day Linux isn't uniquely well-designed or superior as compared to say, OS X. If other modern OSes can get by without a tantrum-throwing diva like Linus, I think it's safe to say it's not a necessary part of the development cycle.