Requiring multiple notaries is a good idea. It solves the problem of how to quickly deal with a notary that's gone offline without introducing a potential for forks. But there are a few problems with this proposal.
The tradeoff is that even one node can stall the whole network.
This is bad.
You can produce a block without a node's signature if that block includes transactions that remove enough votes for that node.
Requiring a substantial portion of the network to be online and do something proactively seems unrealistic. It also takes quite a lot of blockchain storage to get votes from a large fraction of the network.
I'm not sure what exactly to do if a node stops signing and then nobody takes their votes off of that node.
Defeating a notary that censors transactions is really tough. My proposal (in a different thread) solves the issue by taking away the notary's ability to choose which transactions to include, moving that responsibility to nodes randomly selected with proof-of-stake. In order to censor a transaction with my system, the proof-of-stake would have to keep randomly selecting nodes that agree with the censorship, a situation whose probability of persisting over time decays exponentially (unless most of the network actually agrees with the censorship). As soon as nodes are selected that care more about the fee they can claim than the reasons for the censorship, they'll put the transaction in the block. Once that happens, the notary can only accept the outcome and sign the block, or stop producing blocks entirely (which of course will quickly result in the notary's replacement).