This guy explains how to have Meshlab calculate the normals and why, starts to be interesting around 4 min,

https://www.youtube.com/watch?v=kVGENYi4gz8

I think my problem (and maybe yours) is due to the fact that i scanned a rather large model and the scanner does not remain in a fixed position but turns around the model in a vertical plane. So when you tell Meshlab to compute normals with regards to the viewpoint, the problem is there is only one viewpoint for the whole scan and it might not be where your scanner was.

One way to counter this i think is to load each point cloud individually and compute the normals independently for each scan and not for the collection of scans. It takes a bit more time but it probably gets rid of the black spots this way.

My hypothesis...