...
In an inefficient market, the mining rental cost becomes the same as mining revenue, so this scheme almost doubles the attacker's money.
...
I think this is and interestin idea. However, while I find it easy to agree with the quoted assumption when we are speaking about small amounts of hash power, once it's about half the network it becomes dubious. No matter how efficient the market is, if you want to buy half of all the supply you'd be moving the price a lot.
That's why my original argument involves a mojor external event: block reward halving.
I think your idea is interesting, the rental idea is orthogonal and could be combined with it.
I agree that in the current market, renting half of the hashing power for Bitcoin would be really tough. But that is only because the Bitcoin ASIC market is inefficient in the economic sense.
The rental attack idea probably already works - and may have already been used - for GPU mining coins. Even there though there really isnt yet an efficient market for GPU computation - at least not yet.
An efficient market for GPU computation would be the situation where almost everyone runs some smart app on their GPU which is intelligent about earning money through any means - renting out compute time for any of a wide variety of coins/projects/etc when they are asleep/away, but only when the pay is worth the power cost, etc.
I think that market efficiency will tend to increase over time.