I got the impression that the answer you'd get from the Ethereum team is: at the worst case, this is just as bad/good as the current system - i.e. real world effects have to be mediated by an executing authority, same as a real world contract. They would add (and I agree) that smart contracts still have advantages, even in this case.
Don't get me wrong, I'm pretty excited about smart contracts. It just seems like there are a lot of things that we don't think much about currently because our current system is generally pretty forgiving in its fuzziness, but that could have big consequences (and introduce new incentives to act badly) in the world of script-controlled, irreversible contracts.
Oracles will be fine for some use cases. Schelling consensus works for others. There are many ways to get real world information into Ethereum, and many ways to reflect how well real world actors adhere to their commitments.
Somebody (or some other system) has to be in charge of the input to the Ethereum system to tell it that Person A has died. This seems like a huge single point of failure that doesn't exist with current contracts, if it gets hacked then the money gets transferred irreversibly and that's that.
Trying to build a smart contract with any kind of "real world" dependency like this seems equivalent to using a crypto system with an extra key that the government controls, there's a single point of failure that could compromise the whole system.