I'm not talking about going back in time to replace a block that existed in the past to cover up the tracks, then having to catch up with the legit chain. The attack chain would start as just a fork of the existing chain and proceed forward without any handicap. Because it would be a shadow chain, it would not affect the difficulty of the overall network.

I think I understand your solution: The node compares the hash difficulties of blocks in each chain and chooses the one where difficulty sum is higher as that would be a sign that it's a chain from the real world where competition is high. Counting the overall transaction volume and diversity of origin bitcoins would help identify a fake as well.

Legitimate network partition problem will have to be solved. If bitcoin depends on uninterrupted internet, then it could be considered a weakness and may bring a new meaning to words "economic blockade" or "trade embargo".

I wish there was an easy way to find all the past discussions grouped by topic: complexity, block chain, attack, etc.