This is also useful clarification re my subthread with David.  I don't think it's useful to try to conceive an "InfoSpace" for a given XML process or process pipeline which encompasses the entire Web.  My thinking is simply, as above, that node navigation mechanisms are opened up a bit more to resources which at present would be considered external to the document being processed.  I think Michael Kay's key idea was in allowing the XML parser itself to address such resources in a dynamic, late-binding way, more so than the current system of entities and such, and which then opens things up to e.g. XPath largely as it is.  My own idea is to expand beyond that to support callback-type processing within the XML parse phase, triggered according to node patterns, which could support actions such as retrieving resources from the Web to produce additional or modified nodes.  I prefer that separation, and I can also imagine other uses for such a system, including additional and more efficient schema and transform mechanisms.