The issue with a large increase is it creates a slippery slope. Raising the blocksizing is essentially subsidizing transactions. If we persist on doing that someone WILL take advantage of the free space. Now what happens if 8MB blocks get filled up way before the intended increase? It is not an improbable scenario that we could see bigger block get filled surprisingly quickly & the increase will have been more or less for nothing.
miners have an incentive to make smaller blocks as they are transmitted faster and reduce their orphan rate. so they wont include no-fee-transactions forever.
as blockreward gets reduced miners are forced to calculate their cost per transaction (cpu and bandwith wise) they wont allow freebies.
It seems to me you are making two assumptions that do not appear necessarily true to me.
The first is that propagation times are not subject to change (improve). What happens when IBLT is implemented and propagation is constant? There are a handful of other improvements that can decrease propagation time and it WILL get worst as big miners improve connectivity between each other. The incentive to mine smaller blocks then kind of disappears and so does the cost to publish larger blocks to a certain extent.
Second is a most common fallacy that suggest all miners are the same, that their decisions can be projected as a group and not as individuals. That is absolutely wrong. Cost per transactions differ from one miner to the others and bigger miners will have incentive to mine bigger blocks and eventually suffocate smaller miners who cannot keep up with their resources.