We always use NONE filter in our QA environment to test the components. The bugs are filed if problem occurs become of RichFaces code. Unfortunately, sometimes the problems happen because of the JSF implementations (JSF RI and MyFaces), like a bug with &# recoding in the response writer is several versions of JSF-RI. The similar one from the latest MyFaces (http://jira.jboss.com/jira/browse/RF-3557).
Only thing we can do is posting bug to their bug tracking system if it is not there yet. Avoiding such bugs is not always possible.
Thanks -- I see the problem has been fixed. Would never have guessed this would have had an interaction with scripts loading style.
Glad to hear you do QA under NONE.