Fragmentation problems for long running Nim programs made us revisit the existing allocator's algorithms.
It is always turned on.
Allocations over 4KB can be much faster than before.
All GCs written in Nim (refc, v2, markAndSweep) use the new allocator.
Other runtimes are often based on Hoard or jemalloc or similar. They seem to use a less disciplined way of controlling fragmentation and performance issues. TLSF convinced me by its principled approach caring both about theory and practice.