Cool idea, but it seems it would be a pain to implement, particularly if you were trying to make it 030 compatible or some other advanced member of the 68K family. If you don't try to make it compatible with one of those you won't have any software that's designed to run on it.
Also, it seems that the performance gain would be rather small. You're still limited to about 1 MIP or so. It seems it would be better to either implement some 68k compatible chip in FPGA or work on some kind of coldfire accelerator rather than trying to hack on extra instructions to the 68000, but perhaps I'm being overly negative about this.