Incremental only means "construct the DTM incrementally" -- in other words, parse and construct the DTM tree as Xalan actually references it.
If your stylesheet only references the first half of the document, this means only the first half winds up in memory. It also means you don't have to wait for the entire document to be read in before we start processing. It does _not_ filter the document. The entire DTM tree up to the last node actually examined by Xalan will wind up in memory, even if it is on a branch of the document tree which Xalan doesn't examine. Automatically figuring out which parts of the model don't have to be built because they will have no effect other stylesheet is significantly complicated, but is a known research area for future development. Until we get that working, prefiltering the document based on your knowledge of what's needed will have to be left an an exercise for the programmer.
