Performance suffers. Every write requires an adjustment to an object's reference count, and if you have objects that can be cross-thread an atomic test-set at t

Performance suffers. Every write requires an adjustment to an object's reference count, and if you have objects that can be cross-thread an atomic test-set at that. The underlying allocator has to deal with fragmentation, because there is no copying. Furthermore, every object grows a count field.