Things we learned in 2017: in the full-blocks equilibrium, transaction fee prices are even more volatile than cryptocurrency prices. This suggests that for costs incurred over time rather than immediately (eg. storage space), setting a fixed fee may actually be not that bad.
For example, suppose that we claimed that the Ethereum state was "the world hard drive" with a maximum theoretical capacity of 120 terabytes (assume no sharding); you would need to lock up 0.001 ETH to fill up 1 KB of state (now ~$5 for a medium-sized contract).
If ETH went up by 10x, that would make the system less affordable sure, but then *interest* in ethereum going up by 10x would make fees less affordable by a similar amount anway.
Before, I discounted this, because what were essentially political pressures (but which I didn't recognize as such at the time) were successfully keeping fees in the $0.01-0.1 range, lobbying miners' mingasprices down every time ETH (or BTC) went up.
But with full blocks, we don't have that luxury, and we see how txfee markets actually are quite volatile. For expenses that are not in-the-moment in nature, like filling storage, there's no reason why they should cost $0.01 on Tuesday and $3 on Friday.
