Emmsion based Performance Index for Traffic Control Optimization

We do know that optimization can be based on different criteria, and for traffic control optimization (like in ITS, or traffic management) pretty popular becomes minimization of emissions (NOx, CO, etc). So let's focus on such a case:

we do have six lanes arteria ,
we do have demand and supply,
we do have modal split fucntion parametrized (logit, or whatever)
we choose optimization criteria (min of emmision)

and we do need to allocate bus lanes within the arteria.

if we'd base on travel time minimization we'd keep bus lanes uncongested at all times to ensure, that public transport remains attractive. so we'd allocate at least one bus lane within arteria.

in this case private car lanes become jammed.

travel times are minimized, because modal split function keeps on giving more demand to public transport (uncongested).

4 lanes of arteria are completely jammed, and few people inside private cars are stuck in jam. Multinomial Logit Model (say close to reality) behaves the way that there are always some unreasonable agents that prefer to stay in traffic jam then to ride with a bus. Nevetheless most of people have chosen public transport and cumulated travel times are minimized.

but if we choose emission criteria for optimization, than it occurs that those stubborn private car users will generate so much emissions, that it's better for everyone to avoid jams at any cost, even when overall travel times will be much higher.

So how should we behave to keep being rational in this policriterial stochastic world?