I think the issue is that a camera that shoots a very clean image will be able to compress that file more than a camera that captures a noiser image, because compression is most efficient when there isn't a lot of extraneous noise. Each bit of "noise" needs to be stored uniquely in the file, while, for example, a pure white background takes no space to store because the file simply says "replicate this white pixel 200,000 times here"

I hadn't really thought of it in terms of bandwidth, but it makes sense... a few KBytes of savings from each image, times 10,000 images, times x members dowloading adds up to a lot of bandwidth.