[larkery]

Yes - I think you're right that the actual interesting result from NFLT is not that 'optimisation is impossible', but that 'uniform priors are stupid'.