Developing software that can support heterogeneous hardware platforms, adapt to dynamic highly variable environments and meet rapidly changing requirements is a challenge of pervasive computing. Dynamic adaptive software agents are a promising approach to develop software for such environments. We previously proposed an agent architecture which brings together features such as runtime adaptivity, agent mobility, platformindependence, context-awareness and lightweightness which are desirable in pervasive environments. In this paper, we propose and analyse a cost model of network load for our runtime adaptive mobile software agents. We compare our adaptive agents against standard non-adaptive mobile agents using this cost model. Subsequent experimental results demonstrate that our agents outperform non-adaptive mobile agents. We also propose and discuss further features which make our adaptive agent based software solution more suitable for pervasive computing environments.