> too many entries, most of which seem tuned to cheating benchmarks

Even for entries that didn't cheat, the code was sometimes unidiomatic in the sense that "real programmers can write Fortran in any language".

This[0] article articulates the issue with by highlighting an ASP.NET implementation that was faster than more 'honest' Java/Go implementations primarily by not using ASP.NET features, skirting some philosophical line of what it means to use something.

For me, the more interesting discussion of whether a language/library is faster/leaner than another exists in actual idiomatic use. In some languages you are actively sweating over individual allocations; in some you're encouraged to allocate collections and immediately throw them away. Being highly concerned with memory and performance in the latter type of language happens, but is seldom the dominant approach in the larger ecosystem.

[0] https://dusted.codes/how-fast-is-really-aspnet-core