Wireless sensor networks can accommodate multiple applications by using a multi-stack architecture in order to deliver a large number of QoS. Multi-stack architectures can be optimized by allowing packet exchanges between stacks. However, routing loops may appear because of these exchanges. In this paper, we highlight the problem of routing loops generated when the same packet is routed according to two routing protocols. We define the delayable property of routing protocols by considering that some nodes might hold packets in order to avoid loops in the network. We show that minimizing the number of such nodes is an NP-complete problem. Then, we propose two heuristics to address this issue: a centralized deterministic heuristic requiring a global knowledge of the network, and a distributed stochastic heuristic reducing the number of hops from source to destination. Our two heuristics show important benefits: we reach a gain of up to 67% for the first heuristic and of up to 53% for the second heuristic, in terms of number of hops. I. INTRODUCTION Wireless sensor networks (WSNs) are used for monitoring applications. The current trend of WSN is focusing on supporting simultaneously several applications in order to guarantee multiple levels of quality of service (QoS). A mono-stack WSN architecture, composed of a single combination (M, R) of a MAC protocol M and a routing protocol R, cannot be adapted to fit all QoS requirements [1], [2]. However, multi-stack WSN architectures [3] can leverage QoS management. A multi-stack WSN architecture integrates n combinations (M i , R i), for 1 ≤ i ≤ n. All nodes are synchronized: at a given time t, all nodes operate according to one combination. The combinations are periodically activated according to a specific schedule. At the application level, each packet is marked by a label i and processed by the combination (M i , R i). The main drawback of these architectures is the dimensioning of the activity period of each combination. The dynamic dimensioning is not suitable for nodes dealing with limited resources. The static dimensioning may yield to unoptimized performance in terms of end-to-end delay and packet loss. In [4], we overcome this dimensioning issue by proposing a queue-exchange mechanism that allows packets to be exchanged between the different combinations (M i , R i) in order to enhance the network performances (namely, to reduce the end-to-end delay and to increase the throughput). The packet exchange consists in allowing packets marked by label i to be sent with combination (M j , R j), with i = j. The main drawback of this cross-layering technique is that routing loops