How to allocate the parallel inference tasks among the parallel inference units is an important problem of improving the efficiency of parallel inference systems. This paper takes some examinations of three available allocating methods by expriments, proposes a new strategy and, analyses it theoretically and exprimentally.