a bit of a long shot, but i would be curious how the load / latency looks on your nvme drive directly, while running each of the fio tests. (just as a sanity check for drive/hardware issues between scale vs core).
(i know on core gstat will show this data, im not sure how to do it on scale, maybe iostat -x -m 1 100 ?)
on TN core:
gstat -s -p -o -d