Pretty much the same way you do anything else, benchmarking. And no, not a simple "hello world" output, that is a crap kind of test. You need to actually build a realistic site structure in different systems (and hopefully one in raw PHP for a baseline) and then benchmark each of them in optimized/production ready states. A lot of benchmarks simply leave everything as default for development, which means many optimizations that would occur on live systems aren't done and the results are therefore hindered.

Honestly though, benchmarking like this is kinda useless. Build something that works, and optimize as you see issues cropping up. If you can't optimize anymore or the optimizations become unviable to do, do what you'd do anyways and throw more resources at the problem.