Special case: Natural monopolies
One place where a balanced network of trade between classes breaks down is in the tanking monopoly. In the classic dragon example used at the very beginning of this article, the dragon is hitting only one person at a time.
You can have dozens of DPS characters all adding their output to create a single volume of DPS, but tanking is generally not additive on any single enemy. If the dragon can kill the tank with one swipe of a claw, having two tanks doesn't help. The dragon swipes once, kills the first tank, swipes again, and kills the second tank, then goes on to eat the rest of the group.
This creates a natural tanking monopoly and punishes hybrid characters who try to enter the tanking market. In any fight where there is a single enemy to be tanked, all hybrid tanks will be pushed towards their alternate roles as healers or DPS because the monopoly prevents them from competing as tanks. Since this "one enemy, one tank" condition is the default for most MMO fights, it works to devalue hybrid tanking characters across the board.
One solution to this monopoly is to add mechanics which require multiple tanks for key enemies. For example, fights where the primary tank is sometimes teleported away from the fight for 10 seconds would reward teams for taking a hybrid character who could tank for at least a very short period.
In a situation where there is a limited team size, it wouldn't make sense to keep a second tank waiting around the entire fight for such a short window of usefulness, but a hybrid that could both tank and DPS would be desirable. This could also be considered an example of breaking the "steady state" assumption described above.
Special case: Economies of scale
Another issue that can produce market distortions is economies of scale. In a lot of domains, the cost per unit of production is cheaper for large producers than for small producers. This can mean that anything that increases production (say, a piece of equipment that is used for healing, or even an inherent character ability like a healing spell) does more good for a class that already excels in that ability (in this example, healing).
In games this principle often manifests in the synergies that occur when a character is highly specialized. For example, if a DPS character has access to two abilities that each increase their damage by 50%, the net result is a 125% increase in damage because the second multiplier acts upon the first as well. Therefore, a lone character with both these abilities will do more damage than two characters each with a single ability. Even if both a DPS specialist character and a hybrid character have access to identical abilities, they are generally more powerful in the hands of the specialist.
Typically, hybrid characters have their abilities split between multiple areas, and those areas do not reinforce each other. This lack of synergistic effects can make hybrids even less competitive with their specialist parent classes than they might appear at first glance. If it's always easier for a specialist to increase their output, then hybrids will inevitably get progressively squeezed out of the market as characters become more powerful.
Special case: Unique Abilities
One route to making hybrids desirable has been to provide hybrid classes with abilities that do not exist in any of the specialist classes, such as the ability to resurrect their allies in combat or provide powerful unique buffs. However, if all members of the hybrid class have the special ability, it doesn't affect the natural pressure to specialize.
If paladins already have a comparative advantage as healers, that's still true even if their unique buffs are desirable. It may provide a good reason to bring a hybrid instead of a specialist, but the hybrid will essentially be playing a specialist role.
Alternately, if only some members of a hybrid class have the special ability, then it simply shifts the optimal point of specialization. If these unique abilities are too strong, they can be considered as commodities in themselves and thus subject to the exact same specialization pressures as described above.
This is not to say unique class abilities are a bad idea; they've historically proven to be a very useful tool for encouraging players to create teams with a good variety of classes. But they do not remove the basic pressures that push all hybrids to focus on producing the single commodity in which they have a comparative advantage. Thos pressures must be fought by breaking more fundamental assumptions.