On benchmarking
For benchmarks, one small suggestion: you may want to include the CoreMark build provided by Wasm3. Minimal build is standalone (WASI-free). It is still synthetic, of course, but it is a somewhat better baseline than tiny microbenchmarks like Fibonacci.
For a more application-level comparison, I would also suggest trying the existing Brotli, MAL, smallpt, and wasmsynth WASI builds. They tend to exercise a wider range of interpreter behavior and usually reveal more interesting tradeoffs than very small kernels.
None of these are perfect benchmarks either, but together they can give a better picture of where an interpreter is strong or weak.
Anyway, congratulations on Wasmz. Iām very happy to see more serious work in this area.