But from what I heard, to get where it is, actix developer took some really bizzare shortcuts. E.g. hardcoding parts of response. On top of general unsafe usage.
I think the benchmarks are flawed, since they don't account for such "optimizations".
The benchmark is very smart in that sense. It has multiple stages (e.g. plaintext, db access,...). It proves with each stage deeper into your stack. You can disable there parts of your stack (e.g. middlewares, views, controllers, ...) and just operate on low level http request response pairs. And there you can hardcode on top then (as the application). The stages of the benchmark are optimized for testing individual disciplines like the connection and memory management of the low level http server and parser.
In .NET case they changed the language in response to the needs of the Techempower benchmark (and the Unity3d engine) to allow safe and performing ops. .NET also has the unsafe keyword for low level ops and is surely also used. The advantage of .NET is that it was a team with a central manager who pushed all areas, http framework, base class library and language into the same direction. A community like Rust does not have that privilege.
But from what I heard, to get where it is, actix developer took some really bizzare shortcuts. E.g. hardcoding parts of response. On top of general unsafe usage.
I think the benchmarks are flawed, since they don't account for such "optimizations".