There is not a lot of research in this area and I briefly outlined my understanding of the current state when responding to a github issue:
https://github.com/ben-manes/caffeine/issues/345#issuecommen...
https://github.com/ben-manes/caffeine/issues/345#issuecommen...
Makes me wish I could justify adding instrumentation to our 'memoize' wrappers at work, although our caching doesn't really match the use case for something like Ristretto or Caffeine. (Little enough stuff and the latency of a networked cache is fine for us, so "just set aside lots of RAM on some cache servers" works well enough.)