Think this is to do with boundary layers and the fluid frictional resistance component of the total resistance.
The boundary layer at the car body will show laminar flow (as opposed to turbulence). This layer will be almost stationary, right on the body shell surface, compared with the free stream velocity (turbulent depending on speed and viscous forces) experienced the further away. - i.e. - the dirt will be swirling around in the moving air, but is exchanged with the boundary layer air closest to the shell, and hence has time to settle on the bodywork.
The CFD image by Harleybarley shows this very well.