That's exactly what is accounted for in Peter R's math, and it doesn't work. There are many ways to look at why it doesn't work but one way is to recognize that the cost of an orphan comes from the block reward (cost = reward * risk). So without a reward that collapses to zero this incentive no longer applies. You can work out a more complex model where you already have N transactions in the block (with some fees) and transaction N+1 increases the orphan risk (which is what Peter R did), but you still get the same result. It doesn't work.