Problem: Browsers need size of embedded images before they are able
to render the page, if no size markup has been provided in the HTML
In our (small) sample of images, almost half were less than 1K bytes
You might also use HTML size markup to guess which images are small
vs. big
You also might limit the size of the first request, to make sure you
don't get blocked too long if the base HTML document is very long
A browser might ask for the first 1K of each image; even over a 28.8
modem, you'd fetch, statistically, about 5-6 images or at least their meta
data in the first second
Then later, go back for the rest of the images, or maybe the next progressive
GIF/PNG chunk of the image
Don't blindly generate many more requests if your pipe is very full
already; you can't get the requests back once sent, and the user might
scroll down (changing your decision on which images to ask for next)
Theory: as style sheets deploy, the number of very small images will
decline somewhat, since many are logos or buttons often replacable with
style sheet markup