minbucket was erronously interpreted as minbucket + 1 p-values for teststat = "max" were wrong in the uncommon case of zero variances
