Caveat on the error analysis for stereological estimates