Why would you bother supporting both approaches? It's not as if the recursive approach is any more efficient --- and whether or not you start with the recursive approach, you need to keep memory reserved for a worst case queue traversal.
For one thing, because the recursive approach is debugged. If we can show that the new code doesn't change what the old code is doing when that case is not triggered, then then we can be confident that a bug hasn't been introduced.
