When we did trees in SQL, we found this trick very valuable. It makes possible to implement $node->descendants quite efficiently. Node insertion, on the other hand, gets slower, because you have to recalulate the id's (but we only stored finished trees, so we did not care).