everytime the host of the current/next blob o' data goes offline, you need to waste time finding a replacement

Ther is an algorithm in place where the rarest is downloaded. Also a priority list of older peers is implemented. Some protocol also request the chunk from muliple peers (kademlia).