[Bernhard]

Originally Posted by Djub

On the other hand, I can read a lot of papers with refined meshs close to the walls and bluff bodies. How can this be ?

The point for a good LES simulation, is that you want to resolve 80% of the energy spectrum. Near the wall, you will have more energy in the smaller scales, so you need a fine grid there. Only in LES, you want to have cells with a low aspect ratio (which is not necessary in RANS), so you need to refine them in all three directions. If you apply that refinement globally, the amount of cells will get out of hand quite soon. So it is more a compromise to still perform LES.

Be sure to check out the PhD thesis by Eugene de Villiers, (see http://foamcfd.org/resources/theses.html ), he is addressing some of the issues you mention.