The cost of orphaned blocks is very real.
I stated the same. We might disagree on the magnitude.
Since there is no marginal cost in including a transaction to the current block,
let me be more precise:
There is a marginal cost implied by block propagation speed being proportional to size and propagation being proportional to orphan rate. There is also a computation cost of updating the merkle tree and updating miner with it. These marginal costs are today however magnitudes below the lowest non-zero fees paid.
A rational miner will draw the minimum fee policy just above that line.
Allowing for that rational line being above zero, the question is if that rational limit pays for the security we need to sustain. See my previous calc. on transaction fees needed.