Service-based processes represent compositions of software services that need to be properly executed by the resources offered by an IT infrastructure within a company. Due to the dynamic changes in their QoS requirements, service-based processes are constantly evolving and demanding new resources. To ensure agility and support more flexibility, it is common today for enterprises to outsource their service-based processes to cloud environments and recently to cloud federations. The main challenge in this regard is to ensure an optimal allocation of cloud resources to process services overtime. In fact, given the diversity of the resources within a federation and the continuous changes of the process QoS needs, the reallocation of cloud resources to process services may result in high computing costs and an increase in the communication overheads. In this paper, we propose a novel adaptive resource allocation approach which can estimate and optimize the final deployment costs. We use agent-based systems to simulate processes' enactment. To cope with the services' QoS changes and dynamically adapt the initial deployment, we propose an extended version of the Pairwise-Movement Fiduccia-Mattheyses (E-PMFM) partitioning algorithm. Our experimental results highlight the efficiency of E-PMFM algorithm and show that deployment costs are sensitive to the initial deployment and the used partitioning algorithm.