I am very confused about downsampling.
D800 files compared to D3s, D4, D700
e.g. in this review
http://nikonrumors.com/2012/04/10/nikon-d800-vs-d3s-and-d7000-comparison-by-cary-jordan.aspx/
Regardless of everything else, it's probably best to simply ignore that article entirely. He used JPEG images to compare, doesn't know how to measure high frequency detail other than looking at it, did not sharpen the image after down re-sampling, and used Photoshop's Bicubic Sharper to re-sample.
There is software that can measure the high frequency detail accurately. TIFF formatted images should have been used (discussing dynamic range or any sort of artifacts using JPEG images is nonsense). Any image that is re-sampled will necessarily need to be sharpened, hence to only way to get a valid comparison between images if either one has been down sampled is to sharpen both for best effect and then compare them. That is, after all, what you would do if you made two prints and compared them, so it's is the comparison that you want to know results from. And Bicubic Sharper is one of the worst resampling algorithms available.
Does it mean "binning"?
That is, combining data from 4 pixels, and use it like a large pixel (4 times the area).
Binning is one way to downsample, and it has valid uses but primarily it reduces noise. The average noise from 4 sensor locations will have a pretty wide range of values because noise is random, as opposed to the desired signal very likely being much closer from pixel to pixel. When 4 bits are averaged the result will be lower noise and a higher signal to noise ratio. (But it may or may not provide the best edge transitions in terms of color or contrast...)
It probably can suffice to say that at a minimum you would want to downsample using some form of a Lanczos filter. A Robidoux filter is better, but not available for most software.
If you really want to know:
http://www.imagemagick.org/Usage/resize/#resize
That will cure you of asking about resampling images!
On the camera, we can also select the number of Mega pixel we want?
I don't think we gain "binning"/ advantage in High ISO performance.
The camera provides different sensor cropping areas, which are not binning. It also provides different sizes for JPEG images, and while that probably is not actually binning, it is resampling in some way.
