If you have 1/4 of the hashpower today, you will try to remine (fork the chain) when there is a fee of 2R or 50BTC. At 5% of hashrate, this gives a cutoff of 18x the reward for when it becomes possible to backtrack. Traditionally we have always had a pool or two floating at around the 1/4 level. Therefore when the fees approach ~ twice the reward (and they will eventually if the network stays alive) this will become a problem!
I think you are right in all elements but may have slightly misstated this? Would it be better stated that this will be an incentive problem whenever the fees+reward approach ~ twice the average fee+reward?
It seems it is less of a problem of just fees>reward*2 which is expected to become very common indeed.
The assumption was that the average block's fees are much smaller than the mining reward. When the fees become more significant the calculation must include the expected value fees lost. Instead of F>2R, you'll be looking for F>2R+2F_avg. As R approaches zero this inequality becomes easier to satisfy.
We're probably still off by an R or an F here or there. And we haven't even discussed the scenario when R_1 = 2R_2 (when R_1 is the last block to receive 25BTC).