The assertion that a fee market only works if miners act as a cartel is false. The situation in which miners do not act as a cartel simply presents the consumer with a range of prices which they can choose among, paying a premium if they are willing to pay for high-priority or prompt service or taking a discount for making low-priority or slow transactions.
Imagine a market in which there is no cartel. To make it simple, suppose that there are ten miners each with ten percent of the hashing power, and that the block size limit is not routinely reached. Because they are economically rational and facing different prices for bandwidth and electricity in their respective neighborhoods, they all set different minimum-fee policies.
The consumer is faced with ten different price points for a "minimum acceptable" fee, which determines how many of these miners would accept his or her transaction.
So... paying a minimum fee would get your tx accepted by one miner. On average you're going to have to wait ten blocks before that miner gets a block, so your expected tx time is about 100 minutes. Paying a median fee would get your tx accepted by any of five miners. On average you're going to have to wait two blocks before one of those five gets a block, so 20 minutes. Paying the highest fee would get your tx into any block regardless of who mines it, so you'll be in the very next block in around 10 minutes.
The point is that consumers are not faced with a binary "pay enough" or "don't pay anything" choice; they are faced instead with the opportunity to select a level of responsiveness desired and pay for the priority they want or need on a by-transaction basis.
This is a great explanation!
It fits the "unconstrained" block size scenario, which is how Bitcoin has worked for most of its existence, except for a day or so about March 6th, 2013, when the 250KB soft-limit was effective. It does mean that when users create a transaction they have a single-shot at getting the fee right. In the simplified example, if a user pitches their fee so that 2 miners will accept it (out of 10), and then change their mind, deciding that waiting an expected 50 minutes is too long, then they are SOL, the unconfirmed tx can't be changed. Fortunately this is rare.
The "constrained" block size scenario makes necessary the ability for ordinary users to increase the fee. Users will want to update the fee on their unconfirmed tx to manage the instability in confirmation times, otherwise their tx can remain stuck in cyberspace, and they are helpless.
Certainly, protocol block limits should not be hit unless all wallets first support the updating of fees on unconfirmed tx.