Most good news admins set up their servers to time out the binaries groups more quickly. In fact, my favorite news server software (the Cyclone/Typhoon/Breeze series, which has been passed through at least three companies in as many years) sets up news spools based solely on size. When the size limit is reached, it automatically starts overwriting old articles. Makes maintenance a much easier thing.

Also, I doubt that that much backbone traffic is really spent distributing news articles. You have to realize that there are fairly few news servers that actually house those big newsgroups. (Probably one or two for each major ISP, one for each major university, one for each commercial news source -- likely less than a thousand worldwide.) And the distributed nature of NNTP really optimizes everything. I'm sure that all of the real players have one or two big news servers that feed to their customer delivery servers. Of course, the last time I was working on a backbone was several years ago, so I'm probably just talking out of my ass.