Also expand GC interface with "allocate_pointerless". Limit lazy sweeping to the allocation size that is causing the sweep, without adding to fragmentation.
