Our current version of tcmalloc which comes from gperftools. Over time the tcmalloc standalone library which google uses in production has diverged from the version that comes with gperftools.

See: https://google.github.io/tcmalloc/gperftools.html

Modern versions of tcmalloc (the standalone library) have a new mode of operation, "per-CPU caching", where the library maintains memory caches local to individual logical cores on Linux >= 4.18 where restartable sequences are supported and falls back to older per-thread behavior when not supported.

We should evaluate whether or not we could "upgrade" to the standalone tcmalloc to take advantage of per-core caches.