Yes, of course you can cobble something together when availability does not matter at all (it might blow the fuse in your apt, though;)).
I was just saying that in an application with most basic availability-requirements you're not getting the cost down like that.
I.e. even though you could fit that into one rack, nobody actually would (redundancy is measured in powers of >=2). And even though you might find an ISP who won't bitch about you drawing >10 Amps in "half a rack" (cough), you should still be a little concerned about other tenants screwing around in the same rack as your only copy of 80T of data that you care about... ;)
I don't see why you would use 20 machines. I think a good place to start would be RAID 5 using 6 drives so 5 for data and 1 for backup. Which gives 6 machines, assuming your dealing with uncompressed data assuming it's mostly text you can probably use 2 or 3.
How did I screw that up ... woops, let me change it.
I still argue 7 machines though. I mean, sure you could do USB + enclosure or have some more expensive board (with 6 SATA connectors, I don't know how cheap those go). Then you also may need more power, depending on how you use the thing. It's true that fewer machines generally = fewer faults, just as a matter of statistics.
But in reality, in practice, the users will probably have some IBM or SGI solution that is a full-height rack with a bunch of SAS drives or something. I'm sure you've seen those things at trade shows.
But my point here was to try to determine how much it would cost with total baseline OTS hardware.