New and more interesting error correcting bits are going to be needed here. I could imagine a device with 60TB of raw capacity providing 20TB of data you could read reliably though.
WD lists their 2.5TB drive as having "Non-recoverable read errors per bits read" as "<1 in 10^14" and the drive has about 2 x 10^13 bits, so while the chance of an error in one whole disk read is non-trivial, it's not 100%.
Mirroring will still work fine as long as the error is detected. The chance of read errors in the exact same sector of two disks should still be small. Silent errors were always a problem for mirroring.
At what point do we as consumers start considering drives to be defective? Is the underlying ECC suffering because of the need to hit greater storage numbers?
I may be conservative, though. I mean, I also still instinctively gauge program size by how many floppies it would take up . . .
Same with photography. Each raw photo from my 5D II is 25MB. The megapixel race isn't over.
I don't see the rate at which we use space slowing down any time soon, but we have physical magnetic platter limits to worry about going forward.
We will always have room for more space!