Benchmarking shows that mimalloc seems to be faster on real workflows (at least on macOS, still collecting data on Ubuntu). We could switch the default memory pool cases so that mimalloc is preferred.
cc jonkeane apitrou
GitHub Pull Request #10015