In larger projects or in multiproject builds transforming the xdoc files into html may take a significant amount of time.
After looking into the plugin.jelly - especially into the definition of the doc:performJsl tag - I would think that there is an opportunity to distribute the transformation workload over multiple threads/cpus.
The best candidate for doing so would be the for-each loop over docFiles (lines 328-367). It should possible to reduce the overall transformation time with the ant:parallel task.
The effect should be noticeable on multiprocessor build machines and/or builds which transform many xdoc files, as they may for example occur in conjunction with the statcvs-plugin.
Rather than Ant's parallel task, why not use Jelly's thread tag?