Yes, massively. At least with current data, I don’t imagine it would even be possible to measure this on a large scale, especially given the variation in what a computer is actually trying to do. I think it’s made even harder by the fact that software is often targetted at Windows or OSX rather than Linux, so even benchmarking software is near impossible unless you’re writing software which is able to leverage the specific unique features of Linux which make it more opimized.