Richard's analysis is here:
http://tech.groups.yahoo.com/group/F.../message/11264
The benefit of binning on the sensor is that you only get the same read noise for a binned super-pixel as you would get for a single pixel in 1x1 mode. So, for example, with 2x2 binning you are getting 4 times better SNR just considering read noise.
Richard's point is that you also need to consider shot noise and the improvement in that increases as the square root of the number of binned pixels - with 2x2 binning you only get a 2 times better SNR just considering shot noise.
In practice, the benefit is going depend on which of these factors is dominant. For very faint signals the benefit will be closer to 4 times and will decrease to approximately 2 times as the signal increases.
Richard also mentions the problem of saturation of the horizontal shift register which is what causes horizontal blooming on the KAF8300.
Cheers,
Rick.