It appears that the fix for Mozilla bug 406251 (https://bugzilla.mozilla.org/show_bug.cgi?id=406251) is incorrect and is causing regressions in the new Java Plug-In, resulting in the need for further workarounds. At this point, any negative number passed to the INT32_TO_NPVARIANT macro comes through as a large positive number instead of a negative number.
In attempting to remove an earlier workaround for Mozilla bug 406251
related to the handling of large integer values in the NPRuntime, it
was discovered that an earlier regression test case for 6632668 was
currently failing because the fix by Mozilla for 406251 was incorrect.
Added the necessary workarounds to make the test case for 6632668 pass
again. Incorporated the test from follow-on bug 6644253. Mozilla bug
406251 has been reopened.