I'm not sure the rules are being modeled correctly, as several famous configurations like the glider of the F-pentomino do not give the proper results.

I would highly recommend getting rid of the try-catch blocks around the writing, and instead check the bounds yourself. This would increase the speed a lot, and possibly allow the routines to handle boundaries a little more intelligently. Have you run any tests to see how these routines compare to the built in Java ones?