What a disaster. So we've got this data format, XML, that is so inefficient to process that people need fricking quater-million-dollar hardware coprocessors to do it for them???!!! And it's not as if XML is some ancient format that we need to support for legacy reasons - no, it's a relatively recently invented format that could have anticipated this sort of problem.
According to the linked PDF, the XML transformation that the hardware board does is XSLT. A major issue with this language is that it's too easy to write O(n^2) or worse algorithms. You can often convert them to O(n log n) using a feature called "keys", but in my experience this is hard to do and prone to reveal bugs in the XSLT implementation. My guess is that at least some of the customers for this board are suffering from this kind of problem. This is, of course, something that would not be a problem if you used a "real" programming language or database to do the work. Throwing a hardware co-processor at an O(n^2) problem is really not the solution: just employ someone who was paying attention during their undergraduate algorithms course, and get them to replace the XSLT code with something better.