It makes a ton of sense for Impala/Presto, OLAP tends to be about throughput (long queries over lots of data) which benefits from this kind of JIT compilation since the overhead is amortized.
Really excited to see it makes its way into Postgres.
Presto generates Java bytecode which is then turned into machine code by the JVM JIT.
Don't all DB servers "JIT" and then cache query plan?
Executing machine code still can be more performant than interpreting cached execution plan.