Side note: verified that the Speedy Gonzales thing is actually DMA and HDMA updating the MDR, not NMI as first suspected.

http://byuu.org/temp/test_mdrhdma.zip

It brings up entirely new problems by exposing that HDMA is asserting MDR after an opcode's read cycle from which the HDMA triggers; which is going to require a ground-up rethink of the MDR and HDMA sync subsystems. But anyway for right now, even without that, just let H/DMA update the MDR and Speedy will work.