did you try another scheme except of linearUpwind, e.g. vanLeer? Maybe this does not "explode". My general observation is, that upwind for k and omega is fine, though....