Suppose all the miners form a cartel. They will have no problem funding themselves; they can all agree not to include any tx that doesn't pay high fees. The users would pay this fee because they have no other choice.
Some users will try to pay a fee lower than the cartel's threshold. One miner decides to defect from the cartel; he includes in his block all these low-fee transactions. This costs him nothing, so this is a net profit for him (he benefits).
Seeing this, users will know that even if they don't pay the cartel's high fees, they can still get their tx included eventually. Thus, their willingness to pay high fees is lower (that is, the miner consumed their willingness). Thus, more users will try to pay low fees, and the total revenue of all miners decreases.
But it's not just the one miner. Every miner will, individually, have an incentive to include low-fee txs. This means that even with a low fee, it's easy to get a tx to the block. Thus, no user will want to pay high fees, and the total revenue of miners will be low (this is the tragedy - for the miners, and due to the effect on network health, for all Bitcoin users).
This is completely analogous to the classical instance of tragedy of the commons, where all herders would benefit if they all grazed just their fair share, but everyone is incentivized to defect and overgraze, depleting the resource and causing everyone to suffer.
+1 Meni. Great point well made. I think this is a good analysis and description. Of course it could be wrong, but I think its definitely a good framework to think about these issues.
This is not as simple as the classic tragedy of the commons problem, because we also need to consider the impact of faster conformation times on the system. If one miner defects and includes transactions with lower fees, then users with low fee transactions will still need to wait longer for conformations, therefore many users could maintain the high fee level. If a cartel is large enough, it is not clear to me at this stage, if this race to the bottom will occur or not. However I think Menis comments are a very useful illustrative example of a framework to think about the problem, with some sensible assumptions and conclusions, even if some people dont completely agree with the mining cartel logic.
Remember a competitive mining market with many miners is desirable any way, for other reasons. The more competitive mining is the more likely a defection from the cartel is.