The Future of the Web
Step 1: Location Virtualization
The parts are there, they just need to be combined.
Someone will write a BitTorrent extension for Mozilla that posts the pages you visit (a filtered version of your web-browser's cache) onto a BT-esque p2p network. The plugin will also map URLs to BT hash IDs. Thus, URLs may be retrieved from the virtual web network.
When the contents of a URL change, the authentication message for the page, e.g. an md5 hash of its content, will change, effectively causing a flush of the page as an entity in the BT-based virtualized web. Each original page will need to be authenticateable, but there's no other reason to download from the original source.
The effect of this is that you will typically no longer download pages from their original servers. On average, all resources on the virtual web will have the same effective serving bandwidth and latency, near peak and near 0 respectively. Virtual web resources will also be naturally, redundantly backed up.
This is all true so long as the site remains "active" within the browsing population.
Step 2: Attention Caching
Sites that don't stay active will eventually be flushed from the network, reachable only from dedicated hosts. What keeps a site active? Attention.
All you need to host a page is to be connected long enough to upload the first copy of the page. If it's viewed, it goes into someone else's cache. But if it's not, or if it's only copied a couple of times, or infrequently, then it will not make it past the critical threshold of being available often enough to preserve the illusion of immediate access. Resources that do maintain this illusion can effectively be hosted solely by the virtual network. The resource will then be in - what for all intents and purposes is - the conscious attentive memory of the web.
Step 3: Global Semantic Wiki
What is the URL of such a virtual resource? The host portion of a typical URL won't be meaningful in this context, so only the path remains. However, since there is effectively only one host, the path must be globablly unique, and deciding a good global path for an arbitrary resource is probably not realistic.
How about as a page in a virtually shared wiki? Each page in a wiki has a unique name, and is created as a link from another page. To publish a new resource to the virtual web, you would link to it from a page in the virtual wiki. The first edit of the page would leave a copy of your resource (maybe just text, or perhaps a more complex resource, like a music file) in your cache, and as long as you remain online, links inbound to the new page will resolve correctly. If the resource gains attention, it will copy to other caches, and as long as the attention threshold is passed, the resource will remain globally reachable via the shared wiki.
The wiki as a whole will be a gigantic semantic web floating in cyberspace, with no actual locale, its contents ebbing and flowing in and out of existence; a WebMind.