I'm not sure what you mean by "path based indexes", so I can't answer re whether we currently use that particular technique.
Xalan has experimented with a number of approaches to optimizing XPath lookups. We are currently doing some common sub-expression factoring, and under some circumstances we build an index table of elements for fast access by name (though I'm not sure whether we're actually taking advantage of that table right now). If you've got a specific proposal for a technique which might improve XPath resolution speed, then by all means let's discuss it.
