Fission of light actinides has shown to exhibit features, which abruptly deviate from the monotonous trend established for heavier systems. More specifically, the description of their fission excitation functions and fragment angular anisotropies was found impossible to reconcile with the statistical- model framework employing a double-humped fission barrier profile, which was successfully applied to heavy nuclei. This was referred to as the "thorium anomaly"in the 1970's, and was interpreted in terms of the occurrence of a triple-humped barrier. Over the years, its impact on other fission quantities, in particular on the fission yields, has nevertheless been overlooked. In this work we show the drastic influence of the third barrier on the fission-fragment mass (equivalently, charge) and total kinetic energy at excitation energies close to the barrier. The analysis of these observables, and of their variation as a function of initial excitation energy, reveals the suppression of compact (and possibly very elongated) shapes at scission in light actinides. An extension of the semi- empirical GEF code is developed that relates this suppression to the influence of the third barrier. Accounting for this specific feature of light actinides is crucial for a proper description of the fission of the nuclides involved in 235 U-driven reactors, and, to a higher degree, those of the Th-U cycle. These features are not considered in present official nuclear-data tables. Systematic fragment-yield predictions are provided.