(I'm posting this for future reference since I couldn't easily find this practical information.)
I was wondering how much the forward solution changed by using 1 vs multiple integration points per MEG coil. I did a quick simulation with Brainstorm's overlapping spheres head model for a CTF 275-channel MEG system. Here are the results.
(CTF channels are radial gradiometers consisting of 2 coils, so the x axis labels represent 1, 4 and 7 integration points per coil (so 2, 8 and 14 points per channel), as defined in the MNE coil definition file that is also used in Brainstorm.)
The upper tails of these distributions aren't worrisome since they represent a few radial sources (which produce a very small magnetic field outside the head). So the relatively large relative error is expected.