To add to what Shane said, and, for that matter, what you said, something looks indeed to be very wrong. Right now our measurements put Xalan interpretive about 25% slower that XSLTC (in kb/per/sec). The difference should be larger than that, but right now Xalan interpretive is doing some redundent expression elimination that XSLTC is not.
Shane has some benchmarking stuff in the test harness. It probably better to work from these so we can polish the code and have continued regression tests. Paul is really the one to speak on this, since he's been doing a fair amount of work with the benchmarking. -scott
