In the old days of the WWW, AOL users were forced to use HTTP proxies provided by AOL. The setup had the annoying feature of using a different proxy for each request, so that the requests from a single user appeared to come from a large set of very different IP addresses, while each single IP address was used by a large set of users.
Actually, that's probably one of the better ideas of AOL, but of course messy in it's execution. Using all the proxies round-robin, it's easier to even the system load. And even if one fails (and is not automatically offlined from the pool), users will still be able to use the internet (except with the occasional reload of the page required).
When the system was introduced, most of the web pages where static anyway, so the client's IP address didn't matter so much.
But as far as I know, this setup does not violate a single RFC,
Of course. By design, HTTP is a stateless protocol. That's why you have cookies and such.
and websites that can't handle this setup are broken, period.
Yes. And no. For corporate intranet stuff i still require the user to re-login after an IP address change. It's a very complicated technical/political thing. (BOFH-Reasoning: Frankly, i just really hate it when my users carry their running laptops with their spinning harddrives through half the building into a conference room and fiddling with the beamers VGA cable when there is a perfectly good PC already connected and ready to go... which of course always end up not having a network cable plugged in afterwards. But bringing his/her own laptop now serves one purpose less and saves considerable wear on my L.A.R.T. as well as on laptop drives.)
"You have reached the Monastery. All our helpdesk monks are busy at the moment. Please press "1" to instantly donate 10 currency units for a good cause or press "2" to hang up. Or you can dial "12" to get connected directly to second level support."