I would be very happy if we upgraded the P2P protocol with MAC keys and a simple home grown encryption layer, because:
  1. It's practically guaranteed that 5-eyes intelligence agencies are either systematically deanonymising Bitcoin users already (linking transactions to real world identities) or close to succeeding. Peter is correct. Given the way their infrastructure works, encrypting link level traffic would significantly raise the bar to such attacks. Quite possibly to the level where it's deemed unprofitable to continue.

  2. Tor is not a complete solution. The most interesting links to monitor are those from SPV clients connecting to Core nodes. Whilst Java SPV clients have the nice option of an easy bundled Tor client (er, once we fix the last bugs) clients that are not based on bitcoinj would have to use the full-blown Tor client, which is not only a PITA to bundle as Tor is not at all library-fied, but is a giant pile of C which is almost certainly exploitable. Even if it runs in a separate address space, for many platforms this is insufficient as a compromised Tor client could then go ahead and compromise your wallet app too.
Implementing a full Tor client is not a reasonable thing to ask of a wallet developer, but doing HMAC checks and a simple ECDH exchange + AES would be quite realistic.