In this paper, an adaptive matrix multiplication algorithm for dynamic heterogeneous environments is developed and evaluated. Unlike the state-of-the-art approaches, where load balancing is achieved through unequal distribution of the matrix data among the heterogeneous nodes, the matrices in our approach are partitioned into blocks of equal size. Task allocation and the block size are adapted during run time. Data pre-fetch is used to perform efficient communication. Our approach enables the use of various task scheduling heuristics. Further, we show that the control and coordination overheads of this approach are negligible when compared with the overall execution time. The effectiveness of the approach is verified through a configurable simulator developed for understanding the performance of heterogeneous computing environments.