While I love mimalloc, I'm surprised to see jemalloc/tcmalloc not scaling up, and slower than glibc... something is off here!

Not surprising assuming tcmalloc is gperftools tcmalloc. Jemalloc OSS isn’t maintained and neither is gperftools (very old techniques). Tcmalloc has a new OSS variant that’s built using bazel but likely not what's benchmarked here.

Okay, so while https://github.com/jemalloc/jemalloc has been archived recently, there is this fork from Facebook: https://github.com/facebook/jemalloc (7 commits ahead of the archived one so far)

But also this just recently - https://news.ycombinator.com/item?id=44264958

At work we use mimalloc, but a colleague is exploring jemalloc too as he's seeing somewhat better outcomes with one of our app there (Windows)

This paper is above my ability, but as someone who loves Ruby, thank you for working on this Peter and crew!