3x redundancy is very cost ineffective. See this 2010 Google presentation (slide 13) on the 'Colossus' File System, which uses Reed-Solomon encodings.
http://static.googleusercontent.com/media/research.google.co...
There's really a need for a good open-source equivalent to Colossus, which would allow startups like the hypothetical one you describe to do storage must more cheaply and reliably. HDFS (with HDSF-RAID) is probably the closest.. but it's still a long way off.
I'd be interested in what other assumptions you're using to come up with the $5/T profit margin. Are you considering the cost of ingress and egress bandwidth, cooling, personnel to enact repairs? With a large enough fleet you'll have many drives/machines die every day. What about a team of engineers in an on-call rotation to keep to respond when (inevitable) failures occur? Just pointing out that drives + power is only part of the total cost of running such a service.