Especially if they don't have anyone else's beta copy to diff against.
But if they do, it's trivial. Just diff the two copies and put garbage in the differing parts. You'd need to do some checksum thingy to prevent that.
Sure ... another option is to consider all possible pairs of beta testers, and add a watermark that each pair shares. (And possibly also consider groups of 3, 4, etc)
Then if they just nuke what's different, the watermarks that remain identify the exact group. You would need O(n^2) watermarks just to account for pairs, though. (And when you need that many watermarks, it's perhaps more easy to damage watermarks you haven't even found just by writing where they might be stored ...)
If the game is high-profile enough for some sort of large-scale collaboration, another possible problem-scenario is if someone breaks into a beta tester's computer and gets their copy that way; then sharing it with watermarks intact would just point to someone who did nothing wrong except maybe bad security. (And of course, there is the related issue that someone could leak the beta and then claim it was stolen to avoid blame...)