Yet another look at the code tells me that the only way the hwnd can be null after a succesful Window.create() is destroy() or a buffer overrun. Interestingly, Cas' fix for the matrox cards was in fact a buffer overrun fix(*), so your bug may already be fixed.
So Cas, can you upload a CVS build for Egon to test? As an added bonus, he gets to test the NT4 fix at his workplace :-)
(*) Are you not feeling lucky to use java when issues like that arise? I certainly am.
- elias