I'm probably missing something here. CPU/GPU cycles are irreversible, you can't get them back after you have used them. Storage space is reversible, you can get it back at any time by deleting what has been stored. I'm sure they have a way of detecting and preventing this, but wouldn't that imply that constant network connectivity is required to check that the storage is still allocated?
Think of proof of work, where you create loads of hashes for each challenge that comes in, once people came along with ASICs they could create so many more hashes in the same amount of time! The chances of finding a proof with an ASIC is directly correlated with how much more hashes you can create than a CPU in a given time. With proof-of-space you are storing the hashes on disk. The more hashes you have stored the more chance you have of finding a proof. You can delete those hashes at any time if you would like (to use the space for some other purpose) and the only effect is that you will just be lowering your chances of finding a proof for any future challenges. You don't have to be connected to the network all the time, however you will only have a chance of winning Chia while you are connected to the network.
The positives of this approach:
1) Once you have plotted you drive and the plots are farming you use very little electricity (many people farm with Rasberry Pi's)
2) You can utilize unused storage on drives you already own and then as you need that space for some other purpose you can delete the plots.
3) Your chances of wining are directly correlated with your share of the network space ie. someone who plots twice as much has you has on average twice the chance of wining a block, no more, no less.
4) Building a special ASIC for proof of work serves no useful benefit beyond what it was intended to do. Whereas if someone was to build a cheaper more efficient form of storage, well that would be a benefit to everyone!