🚀 go-pugleaf

It seems to me that the moveq technique you mentioned will get you the
fastest code by far.  I may be confused, but here's what I think you were
talking about:
   1) Get the desired screen image (load from disk or whatever).
   2) Write code which will create the 68000 instructions to moveq the
      palette data for each scan line section. 
   3) To display the picture with the correct colors, jump into the
      created instructions.

This will take a lot more work and will be harder to debug, but if you
really want the maximum number of colors/line this seems to me to be the
thing to do.  I used this approach in some digital sampler software I wrote
for the Hippo audio box.  move's and movem's just didn't cut it at 8 to 12
cycles a pop.
This also takes lots of memory, but that's one thing we aren't short on.
                        Dave Baggett
                        arpa: dmb@tis.com