Because the fragment is based on the previous block hash, there is a time limit to how quickly the fragment must be retrieved, thereby proving (after sufficient trials) that the data is physically located within a sphere of radius 10minutes * the speed of light - currently this would prove the data may be physically located on Earth, the Moon and Venus, but no other planet. With a second proof-of-work blockchain established on, say, Pluto, we could then easily prove a similar result for data located on or nearby Pluto. (proving the Pluto proof-of-work blockchain is in fact located on Pluto is left as an exercise for the reader)
I suppose the proof of work could take the previous block hash along with some nonce and then iterate on the data in a 'memory-hard' fashion to add latency. Each nonce+result would be redeemable for a single payment coupon within some fixed time period, e.g. when the next block is found.
I think a good solution to this problem is useful in all sorts of interesting ways... for example, when the blockchain itself is the target data, and the storage payments are transaction fees.