Assuming I understood the method correctly (...), this is because your original mesh is decomposed into voxels (tiny cubes), from which the mesh is re-generated at the end, so the original vertices (and any color bound to them) are likely to be lost. (Probably considered as an acceptable trade-off by
shapeways, as FCS models are just some fraction of orders, and even among those the typical figurine or architecture models are more likely to have an external texture applied to them)
As the automatic "fixing" of thin walls works by inflating the affected voxels, it will easily lead to a "balloon dog" look of the model, so it is probably better to just take the hint from the preview and do any fixing in your modeling software, unless the wall in question is only a tiny fraction below the limit. (Also note that the thin wall check treats everything as an unsupported wall, so you may end up overcorrecting if you take it too seriously.)