You're making assumptions.
"What if" is a game of hit and miss, I don't expect GTX 680 to be slower than HD 7970 but at max core speed it may just trade blows.
We don't know if Kepler solves all heat output issues that are the main limitation in developing the fastest GPU and even though you claim tailoring GPU clock speed will give Nvidia advantage (because, as you say, they're similar clock for clock with the current GTX 500 series), you can just assume
Also going with higher bus speed will increase the cost of production inevitably. AMD can play a price war game at any point.
Possible, but unlikely given that GTX 780 is likely to have anything from 768 to 1024 cuda cores (depending on what is meant by GTX 780, of course).
Clock for clock it will likely be faster.
Going 28nm would reduce heat because current-draw will be lower, resistive impedances will be lower, and hence I^2*R power losses will be lower, but then again quantum tunnelling (i.e. leakage) will be higher, so it remains to be seen where the chips fall exactly (no pun intended), but if I were to bet I think it be in favour of reduced heat, overall.