That's definitely true, technically it could be done, similar to some plagiarism detection software used in academia to tell the level of similarities between publications. But I still think tweaking on the client side would be a more cost-efficient solution... but I agree, they are Google, they should always have free servers and cpu powers to throw in :)
