There are research papers that discuss this. It depends a lot on the bytecode patterns. For Java it's usually as good as the current JITC unless you're relying heavily on auto vectorisation. For Scala it's about a 10% win on open source or a 20% win for enterprise Graal, according to Twitter at least.

For Clojure I don't know. But from what I understand Clojure produces quite inefficient bytecode, so it may benefit a lot.