CFL number
How much, in general, can the CFL number be increased for a steady (implicit) RANS simulation? From test cases, it appears to be around 5. I've used other, similar type solvers where the CFL number can be ramped up to a value on the order of 100.

Depending on the geometry and quality of the mesh, we have been able to ramp the CFL in SU2 to relatively large values. For example, with something simple like the viscous flat plate cases, we have been able to ramp the CFL up to hundreds or thousands. With that said, each application is different, so some experimenting is required.
Have you tried using the CFL ramp options in the configuration file? An example of the format is as follows: % CFL ramp (factor, number of iterations, CFL limit) CFL_RAMP= ( 1.05, 50, 2.0 ) where the first number is a factor, 1.05, that is multiplied by the initial CFL number every 50 iterations until is hits the maximum of 2.0. Finally, an important thing to note is that the stability for implicit calculations is heavily affected by the type of linear solver used and its settings. With the second release of the code, we now have more linear solvers available, such as GMRES. Some the these solvers might offer more stability (i.e. higher CFL numbers) but may also be more computationally expensive. I would recommend trying the different solvers and also experimenting with their settings such as the error tolerance and number of smoothing iterations between each major iteration of the flow solver. 
All times are GMT 4. The time now is 23:32. 