I remember that Chris was thinking about overhauling the rendering of atmosphere fringes some time ago. I think the current system actually isn't that bad-- the transparent "curtain" on the horizon combined with the background sky glow works pretty well for most purposes. There are three main problems with it, but they all strike me as solvable, though this has to be tempered by the confession that I haven't examined the code:
1. The sky glow is a background color at infinity, effectively behind all objects in the universe. That leads to visual errors such as the black new moon obscuring the blue daytime sky of Earth. This could be fixed to some extent by turning the sky glow into a sort of transparent roof, much like the cloud layer, at some cost in performance. I don't think the hit would be huge, but maybe it's not worth it, since Earth's moon is the only really obvious example of this under typical use.
2. There's that weird "hole in the sky" artifact that you see when you're very close to the ground on a planet with atmosphere-- which strikes me as not really inherent to the rendering method; it's more a matter of overzealous clipping, or the equivalent.
3. Oblate planets are currently not supported. If you try giving an oblate planet an atmosphere, it hovers far above the surface at the poles. I imagine doing right by them would be a matter of using somewhat more complicated math to generate the horizon curtain, but I'd guess it could be done.
Personally, I'd vote for oblateness support as the most valuable improvement that could be made. That would make it possible to represent the oblateness of Earth accurately in Celestia (right now it's commented out in the .ssc file), and to give the gas giants more realistic atmosphere models, something I've wanted to play with for a while.