Tokio is fast, but it's huge! Check the size of any library or binary that includes tokio with even minimal features, let alone their recommended default feature set which comes with a complimentary kitchen sink.
If it were dynamically-linked, I'd be more okay with it – but it's statically linked, so every process in memory has its own copy, with slightly different layouts of various structures.
Given how dynamic an async executor is, there's a lot that doesn't get compiled out. Rust doesn't seem to optimise well across dyn boundaries. (I haven't bothered investigating: I usually just rewrite my program to work better, because if I'm encountering this case there's probably other stuff wrong too.)
