Not so much the database as the algorithm; if it's sufficiently understood (as seems to be the case with Google's algorithm, or at least a lot of people claim it to be the case), then spammers can target it directly. Merely having the original data used to train it doesn't give much insight into the algorithm itself.
On second thought, though, being able to identify common characteristics of the least spam-like websites would allow spammers to mimic those characteristics. It would take a lot of effort (figuring out the core bits), but they are clearly willing to put that in. So yes, I suppose that you're right.
