If the instructions were patched on the first use to jump into generated code then the overhead wouldn't be that great & it would run way faster than not being able to run it at all.
Actually, the overhead is rather big. The exception is a minor detail in the overall story.
Just move the following instruction as well in that case, using offsets that will trap if jumped to directly. I'll grant you that it's way more complex at that point.
Ok, then let's make a compuation: Two instructions to be emulated together, so around 2^32 possible combinations. Ok, not exactly, there are less instructions, but you get the picture.
There is another problem, though. The MuRedox trick only works because it can relocate the last 32K of the addressing space to RAM, by means of the MMU. Now, the 68K only has a 24bit address space, and the first 32K are for exception vectors and chip memory, and the last 32K are ROM.
Hence, you'll need a full 32-bit jump, or 6 bytes, not just four bytes.
Anyhow.... If anyone wants to attempt this, go ahead. I even have here a partial solution which emulated a couple of instructions for the first Phoenix core, so you get a full EA decoder already, and an implementation of the bitfield instructions. You "only" have to fill in the rest... (-: