So, to some extent, it's already being done today. Most commonly is a technology called Switched Digital Video deployed by some (mainly Time Warner and Cox) but not all MSOs. What this literally does is only "turn on" channels (generally less popular ones) when they are being watched. If no one is watching them, they are shut off. While this is not over IP, I felt it warranted mentioning as it fits into your goals.
Up next, we already have some MSOs (mainly FiOS) that deliver video on demand over IP, so there is already some desire by MSOs to go there for linear channels, too.
Lastly, AT&T UVerse actually is a video over IP solution (It's essentially video over IP over VDSL2+). There's no reason that traditional MSOs can't go this route (and in fact, many are looking in this direction for the future).
While I would personally love to see the competition from a fully switched network, I'm not certain this will happen. In the case of SDV, if a channel is already on, you will tune the existing channel - so you won't see 10 copies of the same channel switched on for 10 users. This means there is potentially a real infrastructure cost to the owner of the wires if they are required to allow others to use their channels. The same thing with AT&T UVerse - you join an IP multicast stream.
Also, any claims of bandwidth being exhausted on coax are generally false - the problem is the analog channels. On most cable systems, analog channels take up more bandwidth than all digital channels, data, and other services combined. On a completely digital system, built out to 1GHz, there is potentially about 5.5gbps worth of bandwidth available (using current modulation technologies that are already deployed). And this would most likely be limited to just the local HFC node (few hundred houses), so it wouldn't even be shared through that large group of people.
5.5gbps is enough for about 350-550 HD streams simultaneously (depending on if they are MPEG2 or MPEG4, bitrate, etc)