Mathematically this is flawed. Its a data race.
Statistically it may work. Basically it boils down to any attempt to double spend will hopefully be heard before its encoded into the block chain, and recorded. Thus cheating the second transaction out of money. Your maintaining your own internal 'block chain'.
The problem is your percentage of the BTC network's listening node is roughly going to correlate with your ability to detect these events. Also the latency between listening events, and how fast a user can fire a double spend.
Its race conditions all the way down. Your just betting you have the fastest horse.
