Comment by nwh

It's interesting to watch actually, submit a transaction to the network at the moment and there's a rogue node that will mess with the padding of the signatures and rebroadcast it faster than the original. It confuses the reference client into duplicate display, which is what Gox is relying on for the failed/success display. That they're winning races over the normal related transactions isn't that unnatural as the transaction processing stuff has a 100ms sleep() in the middle of it.