these red pixels look more like a bug in VNG implementation to me.
Probably not, because I remember that I saw the same patterns in other demosaicking algorithms I implemented last year for testing purposes. On a quick test, I only found the shown patterns on 135mm focal length shots, but not on 400mm shots. Now what might be the influence of focal length on demosaicing quality?
The 135mm lens is a) used stopped down, thus producing diffraction spikes and b) very nervous to focus and therefore easily produces red halos around bright stars. If the demosaicing algorithms relies on assumptions like smooth color transitions that are well aligned on all color planes, maybe it will fail.
Give me some time to collect more information and find good examples to maybe enhance demosaicing quality (e.g. by combining results of different algorithms).