If the flow and the scalars are not critically coupled with each other and are independent, I suggest solving for just the flow first. Once the flow is converged, freeze it, and then solve for the UDSs. This approach can help to reduce the time per iteration, and hence, the overall simulation time.
Furthermore, once the simulation is stable enough, you can consider reducing the number of iterations per time step. This can also help to speed up the simulation process without sacrificing accuracy.