Actually if you review my prior reply to CfB, then if I am correct that Iota must centralize in order to maintain convergence of consensus (which I am), then the centralized agents will be the ones resubmitting transactions (and redoing the PoW) to put the transactions on a branch that respects the enforced mathematical algorithm.
Thus these centralized agents will need to reimbursed and thus they will end up charging fees.
I am sorry but you and CfB are both wrong.
And proof-of-work is not any more negligible than the actual cost of processing transactions in Bitcoin (for the same hashrate security level). It is rather the centralization that drives up fees due to the power it enables to grab greater profits.
Please don't hype bullshit features.