Scalability - one chain to rule them all is the approach taken by bitcoin (and ethereum)... the solution is many chains + cross chain trading + bitassets.
Light Weight Clients - detecting fraud on light nodes is not the point and he makes it much more complex than necessary. With DPOS any delegate can certify a transaction is valid on behalf of a light weight client. If it is discovered that the transaction was invalid, the signature can be used as proof-positive that the delegate should be fired and in fact could automatically be included in the protocol. In the study of crime prevention, it is not the harshness of the penalty that prevents fraud, it is the certainty of getting caught. Delegates have far more to lose by lying and a light weight client can check with multiple delegates. This completely solves the light-weight client problem and actually opens the door to larger scale blockchains and transaction volumes.
On these two points: barriers to entry is the last thing that keeps large data-intensive blockchains from becoming corrupt, at any time they can be forked and they must compete against smaller chains. They are still subject to their shareholders as well. Thus all of the problems faced by bitcoin/ethereum when it comes to centralization due to Scale of the blockchain go away with DPOS.
Price stability solved by BitUSD... he rambles very badly on these things... attempts to measure supply/demand focus on miners / popularity...
Proof of Stake... identifies problems that have been solved by DPOS... more false arguments about "nothing at stake"...
He didn't address the major hard problem: deriving a random number in a decentralized manner without mining.