This is probably not too hard to make or might even already be possible in the Cobra Engine (but there will be a but). You basically just put an additional camera object in the game (just like the one, through which the player sees the game, you are basically steering a camera object around) and display whatever it shows as a texture. (Simply said.) So your player camera could look at the picture of secondary cameras, which are displayed on game objects. The Unity Engine, for example, can do that as a basic feature.
However, I think it would absolutely murder the performance of the game. Especially once you have multiple cameras, the amount of objects that have to be rendered could go to crazy levels pretty quickly, especially the way some people build.
And from what I read, the performance once your zoo reaches a certain size is already pretty bad.
So while generally cool, I don't think this would be feasible to implement. Just not worth it for the effect.