[Starlevel001]

> all memory could be handed off to a separate dedicated thread that actually calls free().

You could go a bit further and have multiple concurrent threads mark the references, then sweep them up in a separate thread, too. Some sort of concurrent sweep and mark reference count system.

[catlifeonmars]

With concurrency you need atomics or locks to synchronize reference counts between cores, unless you can somehow guarantee that ref counts only decrement and can never increment.