There are ways around this, such as out-of-band signaling, or recipient-provided keys. There are different trade-offs, though, and I haven't yet found the "perfect" one.
I figured you could just scan the blockchain without saving it. So basically find a way to get the daemon to synchronize with the network without saving the blockchain, and get simplewallet to scan those dloaded blocks as the daemon synchronizes.
Of course you can do that but (with some recognition of MKN's laws) does it really make sense to expect someone on a cell phone to receive and scan every transaction being generated by millions or billions of people? Probably not.
My personal favorite for in-person transactions is that the sender signs the transaction, transmits it locally to the recipient via QR/NFC/BT/etc., and the recipient transmits it to the blockchain. In doing so the recipient now knows which transactions require confirmation and scanning. The rest can be ignored. As luigi said though, there are multiple approaches that can work for different use cases.
. Again, I think all thats necessary is a subset of the blockchain to provide adequate ring partners. So regardless of how large the whole blockchain is, you can maintain simply 2 gigs of random blockchain data and your wallet can select from those outputs. This 2 gig subset will refresh and change on some schedule.
