An example I've been wondering about:

Say you have bitcoins at an address and you want the coins to display the following properties (possibly due to output scripts needing to be satisfied):
1.) The coins can be spent, but the pending transaction is broadcast and a specified amount of time must pass before the coins actually leave your wallet.
2.) Sending the transaction again overrides the first, and again renews the pending withdrawal time.
3.) A failsafe address is added that allows the coins to immediately be spent to it.  (very important)

Useful for several reasons:
Transaction mistakes happen all the time by human-error.  This allows users a window of time to change something before it becomes irreversible if they messed up a detail (like the amount, miner fee, or destination)
If a thief obtains the private key to an address and attempts to import/spend the coins, they too are subject to the same constraints that make withdrawals pending for the amount of time the owner originally specified.  Any attempt to spend the coins will notify the owner, and he can send to a failsafe address (which likely has a private key stored elsewhere).  A whole chain of such addresses set up so that the job of hacking 1 wallet becomes the job of hacking many.  
Unlike multi-sig, which just makes hacking many private keys necessary, "pending withdrawals + failsafe" allows the owner to be notified that his private key has been obtained if the thief ever attempts to spend his coins.  

Unfortunately as etotheipi points out, there is currently no functionality that allows successive pending withdrawals to replace each other in the blockchain.  I'm sure that other major changes would need to take place.
Just an interesting idea, I'm wondering if anyone else has comment on it.