An efficient communication strategy is proposed in this paper, which aims to improve the response time and availability of mobile agent based distributed spatial data mining applications. When dealing with decomposed complex data mining tasks or On-Line Analytical Processing (OLAP), mobile agents authorized by the specified user need to coordinate and cooperate with each other by employing given communication method to fulfill the subtasks delegated to them. Agent interactive behavior, e.g. messages passing, intermediate results exchanging and final results merging, must happen after the specified path is determined by executing given routing selection algorithm. Most of algorithms exploited currently run in time that grows approximately quadratic with the size of the input nodes where mobile agents migrate between. In order to gain enhanced communication performance by reducing the execution time of the decision algorithm, we propose an approach to reduce the number of nodes involved in the computation. In practice, hosts in the system are reorganized into groups in terms of the bandwidth between adjacent nodes. Then, we find an optimal node for each group with high bandwidth and powerful computing resources, which is managed by an agent dispatched by agent home node. With that, the communication pattern can be implemented at a higher level of abstraction and contribute to improving the overall performance of mobile agent based distributed spatial data mining applications.