If you spend 100ns you spend 100ns. It doesn't matter if it's in a bottleneck or not. Note that this of course assumes that spending 100ns anywhere is actually making the program and user _wait_ 100ns, i.e. it assumes we are cpu bound all the time.
For any program that does I/O this wouldn't be the case. A "bottleneck" is going to be a network request, and for much of the programs execution time you can just make the CPU do anything without the user even noticing.
So this argument is based on CPU bound interactive programs, which isn't all programs. In programs with e.g. IO (which would be most) then rules 1+2 would come along.
But I guess the thing is that in an interactive CPU-bound program like a game, there _is_ one single bottleneck and most of the code is on that hot path. It's pretty easy to establish that you are in the bottleneck: it's when you are writing per-frame code.
For any program that does I/O this wouldn't be the case. A "bottleneck" is going to be a network request, and for much of the programs execution time you can just make the CPU do anything without the user even noticing.
So this argument is based on CPU bound interactive programs, which isn't all programs. In programs with e.g. IO (which would be most) then rules 1+2 would come along.
But I guess the thing is that in an interactive CPU-bound program like a game, there _is_ one single bottleneck and most of the code is on that hot path. It's pretty easy to establish that you are in the bottleneck: it's when you are writing per-frame code.