I changed the script to only use the first stroke count, which I'll assume is the most common one. I compared the data with the previous version and 51 differences were found with opendiff, all of them appear to be RTK3 kanji. I verified about ten of them and it seems they all needed a correction.
I'm going to update the site with the new data, and assume the rest is correct, so I can move on to other todos.
However I thought I'd post the files on Dropbox ( table_kanjis_update_2012_06.zip 106KB ) so if someone has some free time and is willing to go through it, you can compare both utf8 files with a diff program and see if the new stroke counts are correct.