Other fun glitchy possibilities

Application latitude We noted earlier that applications receive notifications for unretrieved entities. What that application does with the notification is up to it. Mozilla/Netscape 6 handles HTML entities with a local set of descriptions, for instance, rather than pinging the W3C for definitions. Making this tactic more general seems dangerous - among other things, there are no namespaces for entities, just context.
Trying to use documents as external entities The rules for XML documents and XML entities vary slightly in ways that make it sometimes but not always possible to include an XML document as an XML external entity within another XML document. If you want your documents to be used for this purpose, write the XML declaration as an 'encoding declaration', leaving off the Version and Standalone declarations - only specify the character encoding used. Of course, this will cause new interoperability problems if and when XML 1.1 or 2.0 come along.
Node set variability This is a minor problem, but reflects some of the 'do your own thing' built into XML. Parsers aren't required to report content in chunks - they can split it down to character by character, if they like. It's not normally a problem, but it's one that pops up occasionally, especially in processing mixed content (elements and text combined.) It's your responsibility, not the parser's, to watch out for this, and it can make processing a bit more uncertain and switching parsers a bit more difficult.