I've never understood Peter Todd's rationale for promoting replace-by-fee.  

The way Bitcoin was designed, if a node sees a second transaction that spends an output already spent by a previous transaction, it ignores the second transaction.  The result is that the second (and in most cases, fraudulent) transaction propagates to very few nodes and has little chance of being mined into a block.  If the entire network applies this policy homogeneously, double-spending zero-confirm transactions is very difficult, especially if merchants just wait a few second to monitor the propagation of the TX with a well-connected listening node.

What Peter Todd's "replace-by-fee" support does, is allows any unconfirmed transaction to be removed from a node's mempool if a conflicting transaction, received after the original transaction, pays a higher fee.  If all nodes in the network adopted this policy, it would make zero-confirm transactions trivially-easy to double spend.  

Is there some benefit to RBF that I'm missing?  This seems like a reckless policy.