But presumably the computing cost would be much lower for the new coin, since it'd be earlier on in it's mining history. Who knows where the respective expected values stabilize.

Also, what if the new coin has some minor technical improvement? How do we know that won't outweight the established network effects?