The performance impact a programming language can have is rarely shown correctly by benchmarks. Benchmarks usually only demonstrate the performance of different execution contexts.
What if the compiler/interpreter/linker/transpiler (CILT) infrastructure failed to detect certain CPU features when executing code of language X? While the CILT infrastructure for language Y fully identified the CPU capabilities and therefore executes the code faster? What if this behavior does not occur on other machines? Can we trust any benchmark which tries show casing the code execution performance between different programming languages? I think we can not trust such benchmarks.
To me, it is more important to have a good developer experience and a fast feedback loop. If my unoptimized code performs well, well, it's a bonus, but not a hard fact to refer to in the decision phase.