This is nice; I hadn't heard about m-of-n txns earlier.
But on the con side, unlimited, distributed arbitrators may not work out in practice. The more the number of arbitrators, the smaller the intersection set of arbitrators that are trusted by two unrelated parties.
Not with a reputation system. Perhaps the reputation system could even be based directly on the blockchain; just sum the total BTC the arbitrator has arbitrated in the past.
The naive total-arbitrated-BTC solution wouldn't work, since someone can always arbitrate a bunch of 'transactions' between sockpuppet addresses they themselves control. But as long as there's some mechanism to pay arbitrators for their services, it seems reasonable to expect a small set of trusted arbitration services to spring up, analogous to CAs. Of course, once you're paying an arbitration fee on top of a bitcoin transaction fee, it's not clear that you're saving any money relative to a standard credit card transaction.
But on the con side, unlimited, distributed arbitrators may not work out in practice. The more the number of arbitrators, the smaller the intersection set of arbitrators that are trusted by two unrelated parties.