[peterwwillis]

The short answer is that the effective use of the protocol is almost entirely based on the quality of the network of peers involved in distributing the content. For a relatively small internal network that can be quality-controlled by a single team, this may work more effectively than a single centralized service. For the Internet, this is retarded. See PDFs below for some example reasons.

The long answer would have me explaining all of the different use cases of both decentralized and centralized services and a comparison for each. For example, compare bulk file transfer to a domain name service in truly decentralized networks. Keeping domain names requires massive coordination of peers over a gigantic index which can't have conflicts. Bulk file transfer is almost the opposite, as few people have the records (torrent) and the task is mainly in distributing the index (the file). Both are simpler, more efficient, and more reliable if you take out the middle-man of trying to coordinate with more than one peer.

http://www.imd.uni-rostock.de/veroeff/ImprovedBitTorrent_Dan...

https://engineering.purdue.edu/~isl/TON12.pdf

https://www.ietf.org/proceedings/74/slides/P2PRG-0.pdf