That's an interesting hypothesis, but I can't find any source to confirm or contradict it offhand. The part I took the quotes from only mentions that the order of dealloc'ing objects in a collectable object tree is undefined, as is the thread on which dealloc is called. Both of those are easy to keep in mind while implementing dealloc, though. If a resource has to be freed from a particular thread, then dealloc can schedule it to be released on the right thread using GCD. The non-deterministic order of dealloc'ing would rarely be a problem for releasing resources. After all, if a resource is only used via a particular object, and that object is dealloc'ed, then clearly it's okay to release that resource! Perhaps there are complicated cases where resources have to be released in a particular order, but that's no reason to give up RAII for simple cases.