Some other food for thought: there's an oddity with the way that we approach binary optimization in that we care about it a whole lot, but a program will never

Some other food for thought: there's an oddity with the way that we approach binary optimization in that we care about it a whole lot, but a program will never be optimized further than whatever optimizations were in place when the binary hit the user's machine prior to its first use (i.e. the optimizations that were put there by the compiler running on the author's machine or the distro build farm. Why should that be the case? Why should the optimization phase stop there?