https://gist.github.com/gavinandresen/5044482I am not sure the numbers are sound (I think the underlying study that Gavin relied on has some pretty worst case assumptions) but even assuming they are 10x actual cost the idea that there is no cost is simply not supported. A 1 satoshi fee (or even 100 satoshi fee) for all intents and purposes is a no fee transaction.
Let me go with his assumptions implying a rational minimum fee of 0.0008 BTC for a 250 Byte transaction.
That would imply a total fee of 3.2 for 1 MB blocks. We have not even seen that magnitude yet (exceptions were only fucked up transactions).
Means we are not even close to block size limit sqeezing out meaningful fees, so why increase it?
Because the block limit is not creating a fees market. The block reward is too high and masks the theoretical "bidding" process for block space, leaving aside the fact that wallet software does not offer users a way to update a fee on a transaction until it has been dropped from all the mempools.
However 20MB is large enough for a proper fees market to develop. Currently, blocks >900KB are carrying 0.25 BTC in fees. ~5BTC in fees for a full 20MB block and a block reward down to 6.25, means the fees and reward are similar, and the fees market should be healthy.