I'm not sure who can lay claim to the original but "photography is a just a series of compromises".
Noise vs resolution, based on pixel size is a perfect example. Decrease pixel size as long as the noise isn't noticeable under the conditions you want to shoot, to increase the ability to distinguish details. For maximal resolution (e.g. astro as you suggest, presumably limited by wavelength range < pixel dimension) decrease noise as much as possible, by cooling the sensor. Capturing as much of the sky as possible, for a survey for example, needs sheer numbers of pixels coupled with the "right" glass - the LSST has a 2.8 gigapixel sensor array but, I think, only one "lens"
To your questions, I would say that the 20D produces a more detailed, but smaller image than e.g. 1Ds or 5D, based on its pixel spacing and superb noise characteristics. Some of the issues you need to balance can be addressed without resorting to detailed side-by-side testing. However, your plan is the right one because it addresses the whole process and you'll see how you weight the factors : weather sealing vs not, gloved operation in the cold , total pixels vs spatial resolution; weight /size /balance vs features , $$, etc etc. Then to extremes : if you will always desire high ISO, what's your balance between noise and resolution at the temperature you'll be working at. Even though it's a multifactorial issue, it may end up as a no-brainer after you really order your priorities.
Did you check out the 20Da, which was modified for astro work?