The NodeList returned by getChildNodes is actually the node itself which has a cache so that walking down the list linearly, by calling item(i), costs little more than getNextSibling. The liveness aspect of it in this case doesn't really add much burden. However, the NodeList returned by getElementsByTagName also uses a cache, which will be invalidated whenever you modify something in the tree.
So depending on what you're doing you may loose a lot in performance by using NodeList. But, again there is no general answer. -- Arnaud Le Hors - IBM Cupertino, XML Technology Group
