raah, results were corrupted by overflow. Now slice vectorization is
about a +25% speedup which is still nice as i expected zero or even
negative benefit.