Some interesting benchmarks: https://discuss.kotlinlang.org/t/kotlin-native-as-a-go-alternative-in-2021/23665/8?u=darksnake. I guess that the problem is that he uses very small messages and buffers are not filled.

Do you think it might be because of indirection layers, that JIT may not able to finely optimize? According to this benchmark Netty is quite comperable to go - https://stressgrid.com/blog/webserver_benchmark/ (also benchmark of serving static content)

It is a good question to ktor developers. I also think, it should be the same in general. But IO operations should be heavier than the overhead from not best optimization.

I would expect that, the larger the buffers(and transferred files), the less it matters. Also you mentioned few kb buffer size... than probably it's less of a buffer than OS can handle, resulting in higher syscalls frequency.