The massively parallel architecture of the BlueGene/L supercomputer poses a challenge to the efficiency of parallel applications in scientific computing. The specifics of the BlueGene/L communication networks have a strong effect on the performance of the MPI implementation. Various optimizations and specially adapted algorithms cause varying performance results for communication operations. In this paper, we present performance results of various MPI operations for a BlueGene/L system using up to 4,096 nodes. We discuss their efficiency in terms of the communication hardware and investigate influences on the performance of commonly used point-to-point and collective operations. The results give an overview of the efficient usage of the MPI operations and can be used to optimize the performance of parallel applications.