In the collective risk theory literature, horizontal dividend barrier strategies are frequently used to account for a profit participation of shareholders in the insurance business. This is mainly motivated by some optimality results of this strategy available in the classical compound Poisson model and in the diffusion setting. In this paper, we show that the optimality of horizontal barrier strategies does not carry over to Sparre Andersen models in general, by explicitly constructing a counter-example. As a by-product, a heuristic upper bound for the optimal dividend payout in the Erlang(n)-model is derived.