Here is a summary of current developments in the space of decentralised 2-factor Bitcoin wallets. I figured some people here might find it interesting.
There has been very nice progress in the last month or two. Decentralised 2FA wallets run on a desktop/laptop and have a (currently always Android) smartphone app to go with them. Compromise of the wallet requires compromise of both devices.
Alon Muroch and Chris Pacia have made huge progress on "Bitcoin Authenticator", their (HD) wallet app. The desktop side runs on Win/Mac/Linux and the mobile side runs on Android. Sending money from the desktop triggers a push notification to the mobile side, which presents the transaction for confirmation. Additionally the desktop wallet has a variety of other features like OneName integration. It's currently in alpha, but I suspect it will be quite popular once released due to its focus on UI and the simple mobile security model. I've tried it out and it worked fine.
Bitcoin Authenticator uses P2SH/CHECKMULTISIG to provide the 2-factor functionality. However, this has various downsides that are well known: less support for the address type and larger transactions that waste block chain space + result in higher fees.
To solve this problem Christopher Mann and Daniel Loebenberger from Uni Bonn have ported the efficient DSA 2-of-2 signing protocol by MacKenzie and Reiter to ECDSA, and implemented their own desktop/Android wallet app pair showing that it works and has good enough performance. This means that P2SH / CHECKMULTISIG is no longer required for the two factor auth case, and thus it's as cheap as using regular addresses.
Their protocol uses an interesting combination of ECDSA, Paillier homomorphic encryption and some zero knowledge proofs to build a working solution for the 2-of-2 case only. Their app bootstraps from a QR code that includes a TLS public key and IP address of the desktop: the mobile app then connects to it directly, renders the transaction and performs the protocol when the user confirms. The protocol is online, so both devices must be physically present.
Their code is liberally licensed and looks easy to integrate with Alon and Chris' more user focused work, as both projects are built with Android and the latest bitcoinj. If someone is interested, merging Christopher/Daniel's code into the bitcoinj multisig framework would be a useful project, and would make it easier for wallet devs to benefit from this work. I can write a design doc to follow if needed.
Currently, neither of these projects implement support for BIP70, so the screen you see when signing the transaction is hardly user friendly or secure: you just have to trust that the destination address you're paying to isn't tampered with. Support for sending a full payment request between devices is the clear next step once these wallets have obtained a reasonable user base and are stable.