From the pictures, it looks like an issue with the data bus, i.e one of this bits of *all* ram chips are flipping randomly.
Not quite - this is a rev5 board which uses 256Kx1 RAM chips (i.e. each chip has 262144 words, each word being 1 bit, and since we have a 16 bit data bus system 16 such chips are required to complete a RAM bank).
So every single word in RAM is spread across all 16 chips (each RAM chip contributes 1 bit). Since the flaky pixels are 16 pixels apart with perfect periodicity, if it's a RAM problem it's on a single chip.
For that, neither CIA, nor PAULA, nor DENISE is responsible. It is probably a damage in one of the data bit lines between CPU/RAM/custom chips, probably one of the bus drivers between the two.
Display DMA (and DMA in general, i.e. Agnus access to chip RAM) is not behind the bus drivers, only the CPU's access to chip RAM is behind drivers.
This means it's less likely to be a buffers problem but anyway, if it is, it must be a write buffers problem (74LS244) as the routine would be:
CPU reads from ROM (unbuffered) and writes (through 74LS244) to chip RAM the kickstart image, which is then DMA-ed (unbuffered) to Denise for display.
The "easy" test here is to plug in a trapdoor 512 KB RAM expansion and route RAS0 there, thus cutting-off the onboard RAM bank and using the trapdoor as chip RAM. If the visual problems disappear, the fault is with the onboard RAM.
Swapping the low-high write buffers would also reveal if there's a problem there (the vertical lines would shift by 8 pixels).
I'll agree the CIAs/Paula are much less at a chance for fault and the disk problems are more likely caused by the highly likely RAM problems (but again, if it was a drivers problem, this time it would have to be the read buffers 74LS373 as the CPU would have to read malformed data from chip RAM and thus fail to execute the bootblock. Both the LS244 and the LS373 at fault? Much simpler to blame a single RAM chip).
