REPL cluster listeners were enhanced a while back in ISPN-4105 under the assumption that because the value was written locally we could just notify locally instead. Unfortunately this runs into issue when a backup node writes data and the primary doesn't (and then a retry occurs).

We should disable the optimization when using non tx replicated caches. Transactional is fine as we already went through a 2 phase commit and the user would be notified of partial transaction commit.