Nodes can identify themselves by cryptographically linking their IP address with an account (called "hallmarking" in Nxt) which mitigates Sybil attacks. The client can be configured to only download blocks from a subset of nodes linked with known accounts and/or a minimum stake threshold. In an attack scenario this would allow offline nodes to filter out malicious nodes and sync with the proper chain. It's basically an optionally enabled reputation layer for trusting nodes linked to known accounts, not a silver bullet because of the flaws with all reputation systems (long term planned exploit) but still helpful as another layer of security.
You are trusting those with stake to not lie. But remember they can short the coin and destroy the value of their stake and still profit.
We are interested in trustless, decentralized crypto currency. That is what Satoshi pitched to us in his white paper. Satoshi's design is also flawed though.
Besides this does nothing to stop the attack monsterer outlined. Whose stake is valid? Whose is current, the reorganized block chain or the reorganized one? Which one was the reorganized one? You see proof-of-shit is self-referential and thus can't prove anything about itself.