Performance of distributed applications over a wide area network (WAN) has remained an important issue as WAN conditions change over time. Networking community has employed various approaches to improve application performance by adapting to changes in WAN conditions. Composite web services with components distributed over a WAN pose similar performance problems, however, with a different set of challenges and opportunities to improve. With their unique architecture, these composite web services can be partitioned and orchestrated in a decentralized fashion on a distributed infrastructure. In this paper, we investigate how different topologies generated by decentralized orchestration are affected differently by variations in WAN conditions. We present a system that exploits this and adapts to changes in WAN conditions. The system monitors the availability of network bandwidth, evaluates its effect on the performance of each topology using a performance model based on bandwidth availability, and then selects the optimal topology at runtime. The results from our simulation show that the adaptive system improves performance.