Ok, good news on the gaps. Well, kind of anyway.
I ran Kong's test patch, and was pleased to find it was gappy too. Looking down the command queue, this last row of three pixels looked suspicious and I think they give the clue:
No? For a pixel that is 41 steps wide (grid of 41), each centre-point should be 41 steps apart. The first one is 1660, but the next pixel is 1618, which is actually _42_ steps away. Every alternate pixel position is rounded up for some reason, and I suspect it is exactly that, some poorly applied rounding in the controller, converting between the different coordinates systems and scaling them. That's the good news really, if it's in the controller, then it isn't a fix that needs loads of hardware testing.
The bad news is that either way, I don't have much time to work on it and sort out a new release right now, but I'll stay on it and look at it while my heads still in this code.