The JarEntryRevision inner class of XWork's FileManager class lazily decodes URLs by calling replace to change instances of %20 into spaces. Unfortunately, file URLs can and occasionally do contain other % encoded characters. In order for the referenced file to be opened, these % encoded characters must be transformed into their decoded equivalents.
This bug is directly relevant to Tomcat 6, which uses a naming convention of context#subpath.war in its autodeployer to facilitate deployment of a web application into a context like /context/subpath. Tomcat deploys a war named in this manner to webapp/context#subpath, meaning that all absolute file references will contain a #. Because # (along with all other encoded characters except for space) do not get URL decoded by JarEntryRevision's build method, it is impossible to deploy a Struts 2 application named using this convention into Tomcat 6.
I would think that this could be fixed by running the string representation of the URL through java.net.URLDecoder's decode method before handing it to the File constructor. The only snag is that decode expects a character encoding to be passed to it, and I'm not quite sure how to assume the correct encoding in a cross-platform manner. It might be feasible to assume UTF-8 for all URLs.