Benchmarks are benchmarks. You can base your decisions on them, because good numbers on them will give you good numbers on other related things you do, but nobody who has been in this industry for more than a year will take a benchmark as a guarantee those same numbers for your own workloads.

Benchmarking is tricky. The only one that counts is your software running the way you run it. I often run a profiler while running unit/integration tests, but I know the results will not replicate actual use - it's just a proxy, because I don't want to profile everything in production unless the profiler has almost zero cost.