Hi Felix,
Is the use of I+ I- with apparently larger impact on the final structure is not a sufficient justification?
yes, this is true. Did any one doubt it? If you know which atoms are anomalous scatterers and you have I+ and I- then it is a valid idea to refine their f' and f''. However, automatically making decision to refine f' and f'' just based on atom heaviness and presence of I+ and I- seems questionable to me. May be it's just fine but so far I haven't see logic and convincing arguments for doing this.
In structures with good diffraction data and sufficient resolution we frequently see S atoms on the anomalous difference map.
Sure. This is good knowledge sufficient to make decision to refine their f' and f''. Decisions about choice of refinement strategy and model parametrization have to be supported by the experimental data and model quality. In this case you know that your sulfurs are anomalous scatterers and you use this knowledge to decide refining their f' and f''. Pavel