Well written code in any language will almost always outperform poorly written code in another language. Python being 50x slower than C++ doesn't matter if you are using an exponential imlpementation of a linear problem.

The thing, IME, is that someone writing code in C++ is more likely to be aware that they are using an approach with suboptimal O. In a high-level language you can just as easily write a routine with bad O, especially if you don't understand the PL's underlying data structures. So then you get bad algorithm performance + bad Python performance.

That's what I meant. For most cases Go's performance over Python or even some messy C++ offers little use. But in some rare cases that we need to push the very limit of hardware, C++/C/assembly is more preferable than Go.