If there's no fuel mixed with the air (i.e. upstream of injectors), smoother is better but this being a boundary-layer effect, my money would be on "too little difference to measure". N.B. I am referring to what would commonly be described as "surface finish", not "leftover casting flash and grossly excessive corners because your sand-cast core shifted by millimeters and the way your casting core blended into the valve seat sucked to begin with".
Down near the intake valve in a port-fuel-injected application, some roughness can supposedly help with vapourisation (due to having more surface area) and atomisation (due to roughness-induced boundary layer effects) but properly quantified test results are hard to come by.
The highest-performance engines that I've seen the insides of, have 4-valve-per-cylinder intake runners aimed pretty much straight at the back of the intake valve with the port coming in at a roughly 45 degree angle to the cylinder axis, and any roughness left in the surface is too small to be meaningful.
I have seen intake ports in which what would ordinarily be called the short-side radius, has an intentional ledge or ski-jump in order to intentionally break the boundary layer away on that side of the port in a deliberately chosen location, knowing that the bulk of the flow is going to go over the top-side of the port anyhow.