The go memory usage can probably be explained by the fact that go uses fairly large stacks for goroutines (according to https://go.dev/doc/go1.2#stack_size it’s 8kb per goroutine, even though that’s old)
The Rust tokio solution will use tokio's runtime which automatically spawns an appropriate amount of worker threads to parallelize your async tasks (by default it asks the OS how many CPUs there are and creates one such thread for each CPU).
The async-std implementation is roughly similar.