Haven't you read the paper of Peter R about the existence of the fee market without a blocksize limit?
Miners cannot make infinitively big blocks because of orphan risk. Real world constraints are enough to make appear a need for transaction fee.
Ok, let's say you hit the "real world constraints", what then?
Someone will come along, proclaim the new version of fullblockalypse, say the end is near because blocks are at the technological limit and we now have to make people pay fees, etc etc... then they will say they have that fantastic solution of changing the X parameter in the network (like block times issued in 20 minutes instead of 10m to allow larger blocks to get propagated without getting orphaned, or the implementation of some other "hack", to "solve" the "problem") and if we don't do that NOW, it would be a frickin' disaster.
A new rift ensues etc etc and then we have yet another currency after BTC and BTCC.
This is bullshit. It never stops.
Without a blocksize limit fees will rise in relation to the interaction of the limit of the real world environment and the demand, and this is absolutely not a reason to impose a lower capacity on the system for starters.
We are not talking about tweaking the system so he can cope with an increase of usage, we are talking about relieving what prevent the system to fullfill its natural potential. What is non-natural is the blocksize limit.
To say "it's better that we hit a low limit soon rather than take the risk of hitting a very high limit in a lot of time" is nonsensical.
And to deal with the mess created by this artificial limit devs are talking about SegWit or Lightning... who exactly is planning to change A LOT of parameters of the network?