Every time something like this comes up, I always think that we should be working to incorporate resize into the image format itself. Imagine if there was a raster format that was stored such that low-res versions of the full-res image were stored first, and subsequent sections contained more information, progressively building up detail. Then you could choose to send only parts of the file if you knew the image would be displayed in a small area.
Sort of like the old progressive scan jpegs, but with a more formalized method for bandwidth-saving partial transfers.