Interesting
I see this working with an in-line deduplication appliance that stages the data to disk, deduplicates it, then writes out the compressed data and, critically, the deduplication index. It seems like you would lose some raw throughput, but that's probably compensated by not having to write as much data to tape in the first place.