Thanks.
So, there is noise. Lots of noise. It's only 2 hours of integration, and I probably need a minimum of 5 to really do it justice. It's just not easy to get that data, since there are only a handful of dark nights a month, and Orion is really racing towards the western horizon. I hope to gather another three hours at some point, which should help.
The 5D III, as I've tried saying so many times on these forums, is one craptastically noisy sensor! It is NOT, buy today's standards, a low noise sensor, at all. Which is a little sad, for a camera barely three years old. That's where the color noise comes from.
As for electronic noise overall, it's actually fairly low. I used ISO 1600 specifically to get read noise low. Its around 3.x e-. I was also imaging at around 3C (it's the heart of winter here, nights are 15-18F), so the dark current is very low. The reason the darker regions look noisy is they have been very significantly stretched. I had 21.3mg/sq" skies where I imaged this, which is getting pretty close to the darkest possible 22mg/sq" skies on earth. That was necessary to even get a reasonable amount of photons on those dark areas. Still, on a per-sub basis, the darker areas probably only had maybe 5-8 photons/pixel/minute tops!
So, yeah...there is noise. There is always noise, and when you do a ludicrous stretch like I did, that noise can present a bit of a problem. The only solution is to expose long enough to swamp read noise, and integrate more and more. I need three and a half more hours of integration for my minimum, and I would really prefer another 7 hours.
Regarding dark subtraction, you have to match the dark frame temps to the light frame temps. That can be a major PITA, so I stopped bothering and now use dithering instead. Along with Winsorized Sigma Clipping integration, that takes care of the hot and cold pixels, sat tracks, etc. I still use biases and flats, though...and flats actually tend to increase noise a bit as it removes LP and vignetting.