We show that existing methods for primary-backup replication may disrupt the timing behavior of an underlying service to the extent of making it unusable. Furthermore, we prove that this problem is inherent to the standard primary-backup model. The formal proof is based on an analysis of the “local knowledge” available to each party in a correct primary-backup protocol. This negative result implies that entirely new approaches are needed to resolve the problem; on the positive side, the proof offers some hints for designing a solution. © 2012 Alcatel-Lucent.