It would only be beneficial to do the translation if the underlying aarch64 (or x86 or other) could be run in parallel on multiple data elements to begin with.

It would only be beneficial to do the translation if the underlying aarch64 (or x86 or other) could be run in parallel on multiple data elements to begin with. Fuzzing naturally has that property, but certainly there are other uses.