More weirdness: requiring 16 cycles for division (as per Anomie's docs) breaks several things, e.g. triforce pieces geometry in Zelda 3 intro...
This confirms once more that the timing in the driver is pretty sloppy (as if the many games requiring CPU overclock to start would not suffice to prove it)
I might come back to this later, but I'm more focused on understanding why better tile wraparound (needed to improve Tokimeki Memorial ingame graphics) breaks shadows in Zelda 3 intro, and why my tentative implementation of OPT breaks more games than it fixes.
For the time being, I will add a comment about Bof2 in the source.