Holy trinity reigns supreme until you allow more than 5 man groups. Here's why:

Tank-Healer-DPS-DPS has two DPS, tank DPS, and healer DPS.

If you did Tank-Healer-DPS-Support. The Support has to buff himself, the other DPS, the Healer, and the Tank enough to justify that lost full DPS. Sounds simple. But what happens in a raid with Tank-Healer-Healer-DPS-DPS-DPS-DPS-DPS? The amount he buffs that by will go up multiplicatively.

In EQ/EQ2, they DO have support roles. But they utilize a 6 man group. Tank-Healer-Support-x-x-x. I say X because they have hybrids along with DPS. The size of the group and ways support do their role allows them to double up on support. For example a bard and enchanter will both buff each other to DPS level Damage capability.

This is why BRD/MCH buffs and debuffs are somewhat limited. Its to allow for 2x BRD or 2x MCH, or BRD and MCH groups to work. Say a RDM did entirely support. Now imagine if 2x RDM would buff each other (through unique buffs) to do enough buffs to allow them BLM/SMN damage. Now place two RDM in a raid with 2 BLM and a SMN and watch the crazy output.

To put it shortly, group size of 4 doesn't mesh well unless its the only group size available.

Lets explore it a bit further. Lets just say its a consequence we'll just account for. Just up the boss HP on 8-24 man content. Yeah... what happens when you don't get that RDM in the queue for a trial roulette? That consequence we just accounted for just shot us in the foot. We could go further and force a support to show up in each queue to ensure its always there. Well.. thats another way to screw with queue times.

No.. the best balance is to have a support do its role in a way that it always works whether its there or not. And maybe slightly better (by 5-10%) there then not to encourage but not require flexibility in classes.

What I'd like to see is an 8 man group be the standard for dungeons. This would allow support to thrive as a new role. As well as quicken DPS queues at the same time. Killing two birds with one stone.