The development and wide deployment of digital video-based applications have spurred the need of high-performance parallel video processing systems. Clusters of workstations are seen as a technical feasible and economical viable solution on the development of high-performance video processing systems. Based on the results of our previous research efforts, we introduce in this paper a novel predictive multithread NFS client. Throughout an in-depth analysis of our previous I/O performance results, we tune up the main parameters of the proposed mechanism for maximum performance. Our experimental results show that by properly tuning up the I/O system and by taking into account the application profile, a significant reduction on the time required in the processing of video can be achieved.