RE:
How do you abstract and simplify the techniques needed to track external state. For instance, a pool server needs to track who is owed what payouts. Even if an attacker can't compromise the wallet, they could submit bogus payment data to the monitor and trick it into signing away its money. The monitor could potentially store and manage the entire database itself, but there are often better ways. The trick is to make them simple. This is a problem shared with the multi-server approach. Common design patterns would help a lot.
That seems like the hardest problem.
Does the TPM have access to a guaranteed-to-be-unhackable time source? If it does, then a policy like "do not approve more than X bitcoins worth of transactions in timeperiod Y" should work well enough to prevent more than petty theft, assuming you've got some other way of watching the blockchain for unauthorized transactions and shutting down the compromised machine or using an emergency offline backup key to move the funds to a non-compromised wallet.