Reynolds-stress models are recommended by some specialists for strongly-swirling flows, for example those in cyclones.
Certainly, k-epsilon type models often perform poorly when applied to such problems, and to others involving strong body-force influences.
PHOENICS provides a correction to k-epsilon for use when the buoyancy forces are significant; and this is likely to be beneficial for gravity stratified flows.
However, it is models of the two-fluid and multi-fluid kinds which offer the best hope of success when the "sifting" nature of flows with fluctuating body forces has to be simulated.
The turbulence-research community has still not taken a great interest in these models; but PHOENICS provides for them the tools which they will need, when they do so.