What do you mean the SIMPLE did not work? Did it diverge on you?

What you seem to be experiencing is poor convergence and when you add "extra" dissipation it converges. Granted, this is simply based upon what you have posted here. Back in your post from August 17th, you wrote "is good enough i think". These are words that have led many (including myself) astray. If I were you, I would create a region that identifies the poor quality cells. To do this, right-click on Cell Registers and select New->Field Value Variable. Change the type to "Cells Less than" and then pick Ortho Quality under the Mesh variables. A good starting point would be to pick cells with less than 0.05 and select Save/Display. How many bad elements are there, and where are they located? From here there are two approaches. One would be to Refine the mesh in these areas, the other approach would be to use Poor mesh numerics (which basically runs 1st order upwind solver on these cells). To do either of these simply right-click on the created Register in the tree. 

I would not necessarily trust either of the solutions obtained from the above approach, but it would help to identify problem areas.