Mathematically proving delegates are different people isn't necessary.
You realize by saying this you are no longer a "cryptocurrency" and are now akin to Ripple which makes you trust that nodes (aka delegates) won't collude against you.
Incentives work strongly against any attacker. They would have to spend months developing (via subcontracting or doing it themselves) a bunch of features for BitShares to win voter approval to take multiple positions or they would have to buy a massive amount of BTS to be able to vote themselves in, only to then attack themselves by attacking the network. The cost of attack is already very high.
The cost of attacking Bitshares is extremely low.
It has already been done once. How much did it cost "sfinder" to attack? What makes you think it won't happen again? What makes you think others delegates aren't currently deceiving people? How much does it cost to mislead people through deception?
There is nothing necessarily wrong with having a social construct be part of the mix. Lets compare it to massive warehouses filled with asics run by a handful of mining pools. Both methods are worthy experiments and no one knows what POW or DPOS will lead to in the future.
There is absolutely something wrong with introducing a social construct into chain security and then misleading people to think the system is "decentralized" and "Safer than a Swiss bank account." Imo, PoW is evolving into a centralized system but it is still more decentralized than DPoS because it is mathematically provable through PoW that the chain is secured by a verifiable amount of hashpower.
If I want to attack Bitcoin's PoW algo, I have to acquire 51% of the hashpower.
If I want to attack NXT's PoS algo, I have to acquire 51% of the stake.
If I want to attack Bitshare's DPoS algo,
ALL I have to do is
LIE.