I think there is an important difference. When one generator is faced with deciding to take tiny fees and fill up the block or leave them will compare the immediate gain of fees to his future loss via the general reduction of fees. But unless he has significant market share this general influence will affect other generators more than himself.
I was thinking that he should ignore the influence on other generators and simply calculate how much on average he expects fees to rise by keeping the cheapest fees out of his blocks and see if it is worth it. But then I realized that he actually benefits from reducing fees because they will remove marginal generators and decrease difficulty resulting in him generating blocks more quickly. And he will keep these particular small fees from going to a competitor as well.
I still can't say for sure, but I think the best strategy will be to accept all fees.
The bread story is not so good because almost everyone prefers the morning bread and won't wait just for a discount.
This is the problem: there is NO incentive to NOT accept every fee paying transaction, unless the fee is extremely small (ie. it costs more for the ram + cpu time than the value of the fee). Fundamentally, I think there has to be a built in cost for generators to process a fee and this cost must scale with economic activity.
Adjusting the block size automatically is fine, if you can find a function that scales appropriately. This creates an artificial scarcity for transactions, so generators prioritize higher fees.
We need more debate on this issue. This is a serious economic flaw.