Went on a bit of a quixotic quest and made some headway.
TLDR: It's LC2 VRAM. Presumably third party.
---------------------------------
How these simms behave in 16 and 8 bit color respectively. Seems to occur on a 1024 byte boundary. Even mixing a single apple vram and one of these produces similar display.

The VRAM chips used on these (OKI MS514252A) don't support the JEDEC special functions DSF/QSF, which Quadra 605/650 seem to need. This would align with the chips behaving as expected (VRAM testing OK) when accessed via the independent parallel bus interface.
From the Micron MT42C4256 datasheet, used on the working SIMMs:
DSF (pin 22 in SOJ28 form factor) is wired to both the internal VRAM (if applicable) as well as pin 2 on the 68 pin slot. Interestingly, the mystery SIMMs are also wired correctly for DSF, just the chip doesn't support it.
The question I guess becomes, what would be the missing link machines which use 512K VRAM SIMMs yet have an "older" video subsystem? Some research showed the LC2 supports 512K vram simms yet DSF is grounded, so that's our winner.
LC: 256K. 100ns also. DSF (pin 2) is grounded on the connector.
LC2: 256K or 512K: DSF (pin 2) is grounded on the connector.
LC3, Q700, Q900: 256K. 80ns. DSF is implemented on the VRAM connector
Q475, Q605: 512K. 80ns. DSF is implemented on the VRAM connector
LC3:
Q650 devnote:
Q605 devnote:
I find it interesting nothing explicitly called the requirement for the more advanced VRAM, but I suspect there weren't too many third party VRAM simms floating around....?
Anyways, quite enough silliness.