A CMOS chip will allow for software binning, but not hardware binning. So you get some gain by binning your DSLR, but not like in a CCD, where the gain in sensitivity of the superpixel can be almost directly proportional. (And that can help with matching pixel scale, see below).
Smaller pixels = lower sensitivity but higher resolution (ability to discriminate fine detail). So it's a trade off. Yes, higher sensitivity means shorter exposures (well, higher per-pixel SNR, provided you have sampled sufficiently).
A good target is to match your pixel size to your average seeing, allowing for
Nyquist sampling. Beyond that, it's termed 'wasted resolution' since you are sacrificing sensitivity (photon 'bucket' size per pixel). So let's say you estimate that at your site, on a decent night, you get 2 arcsec seeing. Then aim for a pixel resolution of about 1 arcsec.
As a general rule, the longer the focal length, the larger the pixels you'll want. DSLRs have small pixels, so oversample on many setups. In some ways, undersampling is better, since you can use drizzle stacking to recover some of that lost resolution, and still retain your higher per-pixel sensitivity.