Not necessarily, that's an implementation detail. The result of an XPath query is not an array, it's some object that accepts integer indices for lookup.
If you were to return an array, there would be no way to break out of iteration early without paying for every lookup. For something like XPath, breaking out early is a likely scenario.
Additionally composability seems improved, due to all functions being able to be written in a way where they themselves become generators, so you apply all steps only to each result you retrieve. Potentially saving quite a bit of execution time, if you don't need the full collection.