The primary point of P2SH, at least originally, was shorter addresses. With BIP 70 you don't need short addresses any more because the scripts are being moved around in other ways so you could just use regular CHECKMULTISIG outputs. Lately the rationale for P2SH got re-written to be more like "keep the UTXO set slightly smaller at a cost of making the chain larger", but it's unclear if even that will apply if the threshold ECDSA stuff works out.
Anyway. It doesn't matter. Getting BIP 70 widely used will take a long time, maybe years. Until then we just have to suck up the extra complexity from handling P2SH addresses. The code only has to be written once and then it's done. As you can see, it ends up touching a few parts of the wallet.
I am interested in building a RA service as you call it, so would be keen to know when bitcoinj has this implemented.
Right now lots of people are interested in this, it looks like it's going to be a very competitive space. Most people are building their own web wallet to go with it, though running a web wallet comes with a lot of pain points all by itself (a lot of the TREZOR work has been sunk into their myTREZOR web wallet, for example).
I'm not working on the design notes I posted above at the moment. I'm working on regular HD wallets and getting them finalised and launched for real, along with some other things in parallel.