I wrote a little benchmark. it compared reading our stock cxf.xml with Woodstox with reading a fast-infoset blob of it with the most recent (2006?) fast infoset lib.
Woodstox average us: 7486.09 FastInfoset average us: 4825.43 Keep in mind that a typical bus has more XML than this; I just picked the last file we read. This benchmark could be more careful; it makes no attempt to defeat any operating system caching going on. Thoughts? It seems to me that startup time is hard to pass up. The next step here would be the actual infrastructure to make out build create FI objects and to make our runtime consume them in the Spring integration.
