Wasm.fyi tracks compilation, instantiation, first-call and steady-state execution, host–Wasm call overhead, and memory use. It also tracks feature support (by testing actual corpora) and also the performance of each feature.
You can compare runtimes and compiler backends, drill into individual workloads, and inspect the versions, settings, and hardware behind the measurements. The benchmark code and workload sources are public.
If there’s a real-world .wasm workload or a runtime you’d like to see added, I’d love suggestions.