This paper investigates the outputs of abstract logic-based argumentation systems under stable semantics. We delimit the number of stable extensions a system may have. We show that in the best case, an argumentation system infers exactly the common conclusions drawn from the maximal consistent subbases of the original knowledge base. This output corresponds to that returned by a system under the naive semantics. In the worst case, counter-intuitive results are returned. In the intermediary case, the system forgets intuitive conclusions. These two latter cases are due to the use of skewed attack relations. The results show that stable semantics is either useless or unsuitable in logic-based argumentation systems. Finally, we show that under this semantics, argumentation systems may inherit the problems of coherence-based approaches.