A few weeks ago, Jeff Pulver wrote a lament for IP Multicast and the Mbone (an experiment in multimedia broadcasting over the Internet). One problem with IP Multicast (and the Mbone) is it depends upon extra complexity in backbone routers, i.e. in lower layer protocols, in the network. Jeff quotes a friend,
It appears that most ISPs are afraid of the multiplication effect that can be achieved by allowing their customers to send multicast -- and the difficulties this creates for capacity planning and traffic engineering.
One could attribute lack of widespread multicast deployment to its technical complexity, economics or evil intentions, but it's probably just the end-to-end principal, as articulated by Saltzer, Reed and Clark in 1981, at work. In other words, building a broadcast mechanism into the lower layers of the Internet backbone makes less sense than providing the equivalent functionality at the ends -- in this case at many, many ends as we're talking about broadcasting.
The first end-to-end approach to distributing multimedia content was BitTorrent. While BitTorrent is not real time -- it doesn't handle broadcasting as we know it -- BitTorrent did show how to distribute the load of one-to-many distribution among many end points. And as TiVo demonstrates, real time broadcasting isn't always required..
The end-to-end principal is wins again.