I'll have to defer to this paper. It mentions that the core persistent vectors are cache-friendly, but I would suspect that the llvm will always be an upper-bou

I'll have to defer to this paper. It mentions that the core persistent vectors are cache-friendly, but I would suspect that the llvm will always be an upper-bound for the jvm. infoscience.epfl.ch/record/169879/files/RMTrees.pdf‎ I'm very interested in better answers to this question, though.