So, my spare DC/DC psu is even flakier. Took the previous one to my testbench STacy and measured voltages. 5V was waay off spec. Almost 6V
So I put some load on the psu and started adjusting voltages. The trimmers are really crap. Too sensitive and probably just installing it could have changed the value. Note to self; always measure voltages before closing the STacy up.
There are three trimmers on the STacy DC/DC board. Two are labeled 12V and 5V. The third one seems to make no difference to voltages. Gotta investigate that.
This DC/DC is not generating a -12V for some reason. I gotta see what it's even used for.
I put the original DC/DC board back and the low battery light was gone and writing to SD was clearly much more stable. I was able to read and verify 5x 5MB files in Jookie's test without errors for the first time.
So in my case having a clean stable and bang-on +5V has made the single biggest difference in SD performance. So much so that there is now also a clear difference between my cards.
The 2GB Kingston is slower at writes than the 8GB SanDisk and that in turn slower than the 32GB Kingston.
I then ran a comparison with the three drivers with Jookie's test.
PPDRIVER: Potentially the fastest writes but ultimately failed and was not able to complete the test. Needs write verify.
HDDRIVER: Almost as fast as PPDRIVER without write verify. Very respectable speed with write verification but ultimately has occasional write errors and verification has issues.
ICD: Slowest writes (but acceptable) of the bunch with write verification on but the only driver to manage to consistently pass write and verification without issues. The visual cues on the menubar do show constant errors but apparently it retries until successful.
So ICD a clear winner in this use case without a doubt. I've spent hours to reach this conclusion and I'm confident this is true in my case. Obviously another setup on another Atari model might be different but I believe any differences are relative to my findings.