If you could conclusively identify bots, you could not serve them anything and save yourself some work (and bandwidth).

If you can't conclusively identify bots, you'll end up serving 'poisoned' responses to actual users. Doesn't seem like a viable solution.