I narrowed down my collection to 2200 ish "meaningful" games, and in every case had to rename the folder as mentioned above, ie shadowofthebeast3 became "Shadow Of The Beast 3", then copy the folders over to the Pi, and only then ran the script to bulk generate the .uae files. I then did the XML scrape on the "mix" setting, and got, literally, 99% of the pics I needed.
I think it got 2100+ of them, with only a few random wild miss matches - I always always always after scraping use ARRM and sort inname order, then I can look at the columns to establish the name given in the scrape matches the name of the file, so I know its a correct match.
Even with 2000+ games you can check very quickly for miss matches.
I had read that I should avoid the adf files, as the whdl versions had a better "hit" rate, but Ive tried about 20 at random, and 3 didnt work, giving out of memory errors - a worrying fail rate.
I had used the default "empty" uae file, but even when increasing the chip and / or fast mem in that file, it made no difference.
But, if my friends find more that dont work then I will worry - I could just have picked games that have issues by pure chance and its not really an issue.
As everyone else says, I definitely think using whdl versions IS the way to go, especially with multi disk games, as some adf ones insist disk 2 or 3 must be in the "first" drive - a pain......