Wasmtime and wasmer shims are much slower in stress test benchmarks than other runtimes

The results are a bit surprising:

* Wamr: 95 tasks/s
* Wasmedge: 85 tasks/s
* Wasmer: 17 tasks/s
* Wasmtime: 16 tasks/s

The two C/C++ runtimes seem to be much faster than the two Rust runtimes.
At first I though we might be building the shim in debug mode, but IIUC the logs indicate that's not the case. I wonder if there's something else going on here.

I think we should merge this PR and investigate these results separately.

_Originally posted by @jprendes in https://github.com/containerd/runwasi/issues/855#issuecomment-2662561046_