Why do you say it works for only relatively simple queries? When I gave it some thought a while back, I was fairly certain I could generalize the technique for any query where the documents used could be identified at parse time. It's probably most useful for command line processors rather than embedded ones - since they almost always parse and build a document to run a single query. Maybe certain pipeline architectures work this way too? Some XML databases also store the XML as a BLOB and parse them when needed for a query - so I suspect it would work quite well in that circumstance.