Lisp implementations have had excellent GCs that are broadly better than the JVM for years.

Do you have a source for that? I could see a purely functional language creating much less garbage than Java and/or providing much easier escape analysis scenarios, leading to better performance, but I'd be very surprised to see them perform better for similar amounts of heap allocation, given all the work that went into the various Java GC algorithms. I mean, those LISP GCs probably predate wide adoption of threading, right?