Still another issue is the coordinate reference precision, with respect to the buffer distance (1mm precision on a 10cm buffer would result in visibly shifted boundaries). And then there's the number of vertices in the resulting buffer (e.g., 9 vs 19 vs 37 vs 181 vs 361), which influence the best possible approximation to true circle area.
But the issue is most likely either overlap between neighboring buffers or vagaries introduced by a geodetic buffer. You should certainly be able to zoom to the extent of the features which are most unlike your expected area. If there's no overlap, then you need to compare the difference of area values as a fraction of the coordinate reference precision.
- V