Wow, interesting thread.
I am actually working with a truly decentalized global organization (it’s now more like 50 separate organizations connected through organizational protocols): they are creating a centralized intranet site.
However, while the mechanism of the intranet will be centralized (e.g., the server and it’s application), the content and uses of the intranet will remain decentralized.
So, this might be a relevant, non-technology-centric, model for p2p architectures: look at the decentralized (social, content, interaction, identity, etc.) needs of people and implement the most appropriate technical architecture to support that.
Of course, “appropriate architecture” is loaded, in that, if it were a common option to use a p2p network, and the disaster tolerance* of such a network were a factor, I think p2p architectures would be more commonly used.
* disaters could include hardware / service failure, or bad governance, exploitation of user information, etc.