If you unite the geometries, the mesh definitions will be combined also. If you want to have different mesh densities for different parts, you must keep them separate. This will work fine for frequency domain and integral equation but it is more complicated for time domain.
Are you using time domain solver? Are you using TLM meshing? Unless you use TLM meshing, the inner component will be meshed as finely as the outer mesh. I think TLM allows for irregular mesh, but Im not sure. If the object is PEC, it is likely that CST already knows this and skips the calculation for the interior already.