[-Maxim-]

Thank you for detailed explanation, Brian. Sounds like a valid reason to use the k-epsilon model in your case. That makes me want to re-do some old cases where I had convergence problems. But those were steady-state cases where I had to do a transient simulation anyway afterwards, so I did not look into the different turbulence models any further.

Maybe if you have the time, you could write to Ansys about your case and your observations. It might help to improve their software. F. Menter is working on the SBES (Stress Blended Eddy Simulation) model these days as far as I know but he might be interested in cases where the SST model comes to its limits.
Sorry for all the off-topic talk.

[Red Ember]

Well, I think it's a good idea to use ke before SST, although it requires remeshing