I'm mostly guessing here, but when you profile a field, it defaults to doing a mass-weighted profile, such that the value for any bin in the profile is:
Sum (field * mass) / Sum (mass).
If your object is a very flat disk, then there is probably not much mass outside the plane of the disk, so even when you do a spherical profile, the cells in the disk are dominating the average. My guess is that if you switch to doing a volume-weighted profile of the density, you will get answers that differ more. You can do that in the add_field call by adding weight_field="CellVolume".