After the response writers to use wrappers, we are now able to use the supplied encoding where appropriate.
I had successfully tested the use of Chinese characters on Mac OS X (Tomcat and Glassfish with Safari/Firefox/Chrome) using the latest from the trunk and could not reproduce the problem.
I than switched over to Windows. After installing the East Asian character sets, I ran IE/Firefox/Chrome against Glassfish and still could not reproduce the issue. I asked Joanne to help and she was able to have everything working as well using here browsers as well as her test server. At this point, it appears that everything related to character encoding is working as required so I'm marking this as fixed as well as the more specific issue blocked by this.
Work to support the ResponseWriterWrapper should allow us to delegate the encoding detection.