For a client-server based view-dependent rendering system, the overhead of view-dependent rendering and the network latency are major obstacles in achieving interactivity. In this paper, we first present a multiresolution hierarchy traversal management strategy to control the overhead of view-dependent rendering for low-capacity clients. Then we propose a predictive parallel strategy to overcome the network latency for client-server based view-dependent multiresolution rendering systems. Our solution is to make the client process and the server process run in parallel, using the rendering time to cover the network latency. For networks with long round-trip times, we manage to overlap the network latency for one frame with the rendering time for multiple frames. View-parameters prediction is incorporated to make the parallelism of the client and the server feasible. In order to maintain an acceptable view-dependent rendering quality in the network environment, we develop a synchronization mechanism and a dynamic adjustment mechanism to handle the transient network slowdowns and the changes of the network condition. Our experimental results, in comparison with the sequential method, show that our predictive parallel approach can achieve an interactive frame rate while keeping an acceptable rendering quality for large triangle models over networks with relatively long round-trip times.