[Shiraz (Ray)]

Because software binning (used in the 1600) forces data from (say) 4 pixels into one pixel, it increases the pixel-level SNR by 2x. But then your software will resample the binned RGB to blow it back up to the scale of the Lum for LRGB combination - so you get back sort of to where you started, although fine RGB detail will have been lost and the RGB will be a lot less noisy. Since the binning/resampling process reduces noise, you could get by with lower quality (quicker) RGB data.

However, filtering can also get rid of some noise, so it is probably just as effective to run a smoothing filter over the RGB to reduce the noise.

To check that I haven't got this wrong, I just processed an image that started out with noise of ~340ADU. Binned it 2x2 to give the (expected) noise of ~ 170ADU. Then resampled with the Lanczos algorithm to get back to the original image scale. The final image had less detail, but the noise was ~155ADU - so the binning/resampling process certainly improved the SNR. However, I got a similar looking result, but with noise of only ~130ADU by running a 3x3 median filter over the original image. ie, Don't bother with software binning to improve RGB SNR - just filter it.

Your overall strategy of trying to get by with less detailed RGB is helpful in some circumstances. I just processed an image where I had many hours of luminance, but the clouds meant that I could only get 12 minutes each of RGB. The RGB was very noisy but, with heavy filtering before combining with the L, the LRGB image came out quite well.