It's a cool idea, but the way it tightly couples itself to a model, forcing db calls for something like this kinda ties your hands.
For high-content sites like blogs for example, the idea that a template tag would kick out to the db for every post on the page is too much.
Instead, it'd be nice to decouple the bot detection from the gibberish generation so you could do useful things like precache the gibberish in Redis or on your content object directly and then just do {% if is_bot %}{{ post.gibberish }}{% else %}{{ post.content }}