Original images often are very large in file size (some MB) even if width and height are not so big.
OpenWGA generated derivates in (fixed defined) smaller sizes but uses the original as "fallback" if all generated derivates are to small.
We now generate an additional derivate in the original width and height. This derivate typically has much less bytes then the original file.
We also changed the scoring when looking for the "best" derivate for a given width/height: if two derivates have the same score we now take the one that has less bytes to be transferred (until now we take the bigger one in this case).