Thanks, Roger.

Having thought about it more overnight, and looking at some data dumps I took before repacking the drive for return, the evidence points strongly at the Yansen SSD itself. I see large swaths of good data on the drive, interspersed with the odd 32-byte chunk here and there that has 4-8 bytes in a row of WRONG data.

This suggests the drive firmware is buggy in PIO modes. When I write to it from the PC, UDMA is used instead of PIO, and that works just fine (if not a bit slow).

So.. Run away quickly from "Yansen" brand SSDs.
And avoid any that have the SM2236XT controller chip.

Back to hunt for real Kingspec drives now..

EDIT: Found a 1.8" form factor with 44-pin IDE/PATA connector: https://www.amazon.ca/gp/product/B0099395FA/
Ordered, should be here tomorrow. We'll see if it works correctly or not.

Otherwise, modern fast CF Cards remain a good alternative,
as does Roger's mSATA solution.

EDIT: And for that matter, even these suspect Yansen drives might still work just fine so long as all of the formatting and file copying is done from a PC. The empeg itself would still have to build/write the database files, but from the dumps I looked at the corruption only seemed to happen after a long stretch of sustained writing. So the relatively small database files might write out just fine.

I wonder if this has anything to do with the misaligned partition table (SSDs very much prefer 4096 byte alignments, whereas default partition tables of the era only gave 512 byte alignment).

So if I had some of the Yansen drives, and couldn't RMA them, that's how I'd treat them: do all of the setup from a PC.