Hi Sander,
Your white reference looks correct; it has selected basically all of the stars in your Preview03. It seems there's also some faint nebulosity selected, but I may be wrong. Anyway the selected stars should provide enough statistical basis to compute reasonable white scaling ratios.
Your background reference can be improved. More background pixels should be selected, ideally as large solid regions instead of isolated small structures. You're probably using too restrictive selection parameters. Again though, it should work well to sample the background, but it can be improved.
When possible, I recommend to define more previews and David's PreviewAggregator script to gather them as unique reference images. This holds for both the background and white references. More is always better here: more stars provide better statistical support to implement the star-based white balancing paradigm, and a large number of sampled background pixels provide the correct reference to apply an accurate color correction.
