Although, thinking a bit more on this (and your more helpful reply to another poster), it might very well be that for web sockets, a better performing solution for python might be to write a wrapper for uWebSockets (the c++ implementation) in python.

I'm guessing that the scaffolding code for the websocket-part in python when working with uvloop might indeed give a meaningful performance and/or memory hit (I'm leaning towards memory probably being the most significant difference here).

I'd be interesting to compare a python+uvloop websocket implementation and uWS (both using nodejs and c++) -- and at some point see if wrapping uWS for python would make a meaningful difference.