This example says that you need to satisfy a multisig with two keys, until the transaction reaches a certain depth, when it can be redeemed by one script. The idea is that it removes the risk of the counterparty disappearing by giving the first guy the ability to reclaim his money in the future.
To be clear the protocol implemented now is that a time-locked refund tx is created when tx1 is created that returns the money in the future, but ultimately there will always be a short time window where the counter-party can vanish leaving the funds locked. Jeremy Spilman's solution was to have the counter-party also include some funds, so that their funds would be locked as well so the counter-party has strong incentives to not allow this to happen; I'm not sure what bitcoinj has implemented.
How can the counter-party vanishing affect this? You get the refund (Tx2) signed by the payee before you (payor) sign the funding transaction (Tx1). That essentially makes the whole operation atomic: either all parties get their money into an escrow that automatically returns to the payor if one of the parties disappears... or nothing happens.
