Nicholas,
The reason for this is why BGP is different from other routing protocols. In fact, it is legitimate to view BGP not as a routing protocol at all, but instead as an application that supplies you with what’s called NLRI (Network Layer Reachability Information). Think of it this way: While routing protocols tell you HOW to get somewhere (they are always concerned with the next hop on a hop-by-hop basis), BGP tells you WHERE you need to go, but not how. It is up to the router to figure out how to get to the destination supplied by BGP, by looking at its own routing table / CEF to figure out the next hop.
With this BGP philosophy in mind, by not changing to Next Hop Self, BGP allows your internal routing protocols to make the determination what is the best exit point in your internal network to get to the Next Hop being advertised by BGP. Of course, this assumes that your IGP knows about the link in question.
As further evidence of this philosophy, look at the Administrative Distance of internal BGP: 200! Why is it so high? It is because the designers of BGP want iBGP to be the last resort (compared to any true internal routing protocol) for internal routing purposes.