Picard does not use the ISRC from disc but applies metadata from MB to your files, so I don’t think this is a bug. Reading and using ISRC from disc if data was loaded via Disc ID lookup could be a feature, though. Currently Picard intentionally does not read ISRCs from the CD when you perform a disc lookup, since reading ISRCs is way slower then just reading the TOC required for the disc ID.
If this is actually a different recording it should be split out. But this is difficult to tell without having access to this release and other releases which contain the recording. From what I can tell from the data we currently have there is no strong indication that it is different. A two second difference usually doesn’t tell much, it’s pretty common that the same recording has e.g. more or less silence at the end depending on the CD mastering. But in case you have access to the Crossroads track you can probably compare it to the track from the other disc you are looking up.
Actually what is listed on this page is from MusicBrainz’ perspective not “many versions of a song” but the very same recording. If there are different versions in this list they should be split out as their own recording (see also above).
In this case I don’t think there is much we can do currently. The recording has multiple ISRCs assigned right now. Whether the track on Crossroad got assigned the wrong ISRC by accident is hard to tell, that’s the labels business.
What I think is missing for MusicBrainz is to track from which release a ISRC originated. Then you could know which ISRC applies to your specific release version. See also the related discussion at: