Are there any improvements in there that would allow for more efficient bytecode interpreters in WASM?

As far as I understand, something like e.g. a high-performance JVM would be hard to do in WASM due to its machine's Harvard-like architecture.