Traditional mobile agent (MA) migration mechanisms focus mainly on asynchronous distributed system, and thus encapsulate the whole migration process into an inseparable action. The lack of consideration of time-space consistency makes mobile agent system (MAS) necessary to be under the control of certain high level architecture (like HLA) in distributed simulation systems (DSS). However, the scheme imports migration latency which is unbearable to those systems with strict restrains on real-time performance. By utilizing the decision ability of the task execution agent, the migration granularity can be decided based on the exact task. Then the better efficient migration scheme can be chosen to meet the synchronous requirement and then decrease the migration latency. Contrast experiment is done and the new mechanism performs better time-efficiency in most conditions. This means that the intelligence of mobile agents can contribute to the migration mechanism details and boost the general performance.