See my other comment for details, but in short, this strategy uses a single code base for both interpreter and JIT. So any further performance improvement will benefit both without any additional work. The traditional JIT-only approach is costly to maintain in comparison.