I guess that establishes that the 2nd 'page' is a pixel higher in VRAM...

it could be a 'bug' in the hardware whereby once the scroll counter exceeds a certain value it ends up pulling from a line higher I guess?

or a bug in the CPU emulation when it's writing this data in the first place since we now know it's software rendered...

the videos I can find that look most like real hardware aren't doing it.

is that method of scrolling with a bitmap layer known to be used anywhere else?

http://mamedev.emulab.it/haze/