In a wide area multi-cluster and Grid environment, the systems usually have a deep hierarchical network organization with highly unbalance bandwidth among sites. In order to communicate efficiently in such environment, a multicast tree must be constructed dynamically after the parallel tasks have been assigned to a group of processors or nodes. A new heuristic algorithm named GADT (Genetics Algorithm based Dynamics Tree) has been proposed to efficiently find a near optimal multicast tree. The results clearly show that GADT provides a good performance when compared with the binomial tree algorithm and optimal algorithm using exhaustive search. GADT is an attractive algorithm for generating a broadcast tree for long running applications. Hence, GADT can be well applied to speed up any MPI implementation including MPITH.