Let me explain by example: 10000 users seed the same 10 GB file, if they will divide it, taking into account redundancy - 20 times multiplication, for example, each user will have only 20 MB piece.
In my opinion it can be implemented just as a client feature, not a protocol modification: technically clients should automatically remove superfluous pieces of torrents, based on DHT information, I suppose. But Im not quite sure.
It should be used for data what is not for everyday use: disc images, distributions, installers, movies - stuff that is quite common in bittorrent. But of course user should decide for what torrents this feature will be enabled.
This should just make seeding much easier.