Ok, I have done a couple of tests (repeated so as not to be left with a single result and but with it I can average)
I reduced the list of prefixes to 59 (before they were more than a million) and also shortened the prefixes, to 7 letters (before they were 9 letters) and with this if I can get results only with the GPU
the query was this: (Without CPU)
vanitysearch.exe -b -gpu -t 0 -o results2.txt -g 2424,128 -i testShort.txt and the results were: 65
with CPU, the query was:
vanitysearch.exe -b -gpu -t 8 -o results2.txt -g 2424,128 -i testShort.txt and the results were: 126
I do not know what to think,
Perhaps as you said, it is not yet well Optimized for generation 30 cards and therefore, it "depends" a lot on the cpu for a long list and / or Long prefixes.