The two-machine flow shop problem with buffers where one of the machines has constant processing times is studied. It is investigated if optimal permutation schedules exist, i.e., schedules which minimize the makespan within the set of all schedules and where the job sequence on both machines is the same. Two standard buffer models (an intermediate buffer that a job occupies between its processing on the two machines and a spanning buffer which is occupied by a job over all of its processing steps) are considered for two types of buffer usage: i) all jobs occupy the same buffer amount and ii) for each job the occupied buffer amount equals its processing time on the machine with non-constant times. For (i) and both buffer models, it is shown that the set of optimal schedules always contains at least one permutation schedule. For (ii) the existence of optimal permutation schedules is only guaranteed for instances with n jobs for all n≤6 (spanning buffer) and for all n≤3 (intermediate buffer). For each n>6 (spanning buffer) and each n>3 (intermediate buffer) exists an instance which does not have an optimal permutation schedule. For (ii) and both buffer models, an upper bound and a lower bound for ratio between the makespan of a best permutation schedule and the makespan of an optimal schedule is given. It is also shown for (ii) and both buffer models that in general it is NP-hard to decide if an optimal permutation schedule exists.