This possible issue is related to the following fix:

6766644 2/4 Redefinition of compiled method fails with assertion
            "Can not load classes with the Compiler thread"

that was integrated in HSX-21-B02.
    
On 3/18/2011 12:06 PM, Tom Rodriguez wrote:
> By the way, I think there's another theoretical issue with the
> compiled method load events.  nmethods act as weak roots during
> full gcs so there's nothing stopping a locked nmethod having the
> oops it references get unloaded before it posts the compiled method
> load.  The easiest fix is having the nmethods referenced from the
> service thread processed during the strong roots marking phase, in
> the same way that nmethods that have live activations are processed
> during this phase.  It's a pretty theoretical issue but it occurred
> to me while thinking about other things.
>
> tom