Nowadays, many large-scale scientific and engineering applications are usually constructed as dependent task graphs, or called workflows, for describing complex interrelated computation and communication among constituent software modules or programs. Therefore, scheduling workflows efficiently becomes an important issue in modern parallel computing environments, such as cluster, grid, and cloud. Task clustering is one of the major categories of task graph scheduling approaches, aiming at reducing inter-task communication costs. In this paper, we propose three new task clustering approaches, Critical Path Clustering Heuristic (CPCH), Larger Edge First Heuristic (LEFH), and Critical Child First Heuristic (CCFH), which are expected to achieve better task graph execution performance by trying to minimize the communication costs along execution paths. The proposed schemes were evaluated with a series of simulation experiments and compared to a typical clustering based task graph scheduling approach in the literature. The experimental results indicate that the proposed CPCH, LEFH, and CCFH heuristics outperform the typical scheme significantly, up to 21% performance improvement in terms of average makespan for workflows of large Communication-to-Computation Ratio (CCR).