A good shape can perform well when rough. A bad shape when rough will be even worse.
Link above explains it better than I can. I remember some wind tunnel experimenting from school we did that was based on old NACA/NASA research into boundary layers and turbulent flow. Rough finishes are never 'better' on their own, but have a nearly insignificant impact on 'good' shapes but will make a significant difference on poor shapes. The right shape that is also smooth is always 'best' but rarely is the 'right' shape achieveable or best for the entire operating range and so roughness or less-than-deal geometry is often used to expand the operating envelope to suit the application. Turbulators, winglets, or other surface flow 'energizers' are often employed for this reason.
I'm betting the reason rougher finishes seem to out-perform smooth ones is because the rougher finsh will make poorly shaped areas easier to detect and address. As with everything, it's likely not the rougher finish in isolation making the difference in performance.