A series of experiments, performed on plant models with ultra high dilutions (UHD) of arsenic trioxide at 45th decimal potency has been reviewed with a particular focus on variability. The working variables considered are: the number of germinated seeds out of a fixed set of 33, the stem length of wheat seedlings and the number of necrotic lesions in tobacco leaf disks inoculated with tobacco mosaic virus (TMV). A thorough comparison between treatment and control group has been proposed, considering the two main sources of variability in each series of experiments: variability within and between experiments. In treated groups, a systematic decrease in variability between-experiments, as well as a general decrease, with very few exceptions, in variability within experiments has been observed with respect to control. Variability is traditionally considered as control parameter of model systems. Our hypothesis, based on experimental evidences, proposes a new role of variability as a target of UHD action. This hypothesis may help interpret unanswered questions that keep rising in basic and clinical research in homeopathy.