Am I missing something? I can't think of a single way to design a system where it is economically advantageous to move away from centralization (where we are intentionally neglecting the possibility that bitcoins might be worth more in a more decentralized environment).
Heat dissipation, possibly.
A huge mining datacenter will produce a lot of heat, and may not be able to make efficient use of it.
Whereas, a million individuals with ASICs in their hot water heater, may be able to utilize close to 100% of the heat. Then their marginal cost to run the miner is close to zero. The difficulty would rise to make the marginal revenue close to zero also, but it would still be an advantage to decentralized mining.
That's a really good point. The amortized cost per hash, η, in the miner's profit equation can be made very low if the heat byproduct from hashing is useable:
I think this means that a miner who has a larger η value (e.g., one who can't use the byproduct heat efficiently) must compete by having better connectivity to network hash power so that he can produce larger blocks to claim more fee revenue.
It would be nice to make a table of the aspects of mining that favour centralization versus those that favour decentralization.