in reply to
perl implementation of Fast Infoset, or other binary XML representations?
My first suggestion: you could give modperl a try.
With modperl you are able to parse the xml file once, when apache starts up,
and store the results in a persistent data structure,
accessible to all threads/processes which handle the cgi requests.
If that's not what you are looking for,
There would also be the possiblity of using a xml database.