I don't like the idea of reversing-transaction, but your idea isn't too bad. But i think there are few problems such as :
1. Your idea require new proposal and hard-fork.
2. This can be abused to tricking seller, merchants or the receiver in general.
3. AFAIK, your idea could be achieved similarly with scripting, specifically with HLTC.
4. Increasing node loads since the node need to do extra checking.
5. Even if majority accept the idea, i'm sure the adoption will be very slow and another major chain-split will happen.
6. Your idea is killing zero-confirmation which is needed if we ever want too see Bitcoin as payment option. I know zero-confirmation is risky, but it should be fine for small transaction and the receiver can re-broadcast the transaction to different nodes to prevent double-spend attempt.