Abstract The abundance of cloud resources has enabled not only web applications, but also scientific applications to easily scale to meet their objectives, such as performance and costs. However, due to the complex and large-scale nature of scientific workflows, the decision on such scaling (resource management) is much complicated often resulting in inefficient use of resources. In this paper, we present RDAS+ as a resource demand aware scheduling algorithm to optimize resource efficiency for the execution of scientific workflows in clouds. RDAS+ maximizes resource utilization by allocating the minimum number of resources (virtual machines or VMs in clouds) with little sacrifice of completion time (makespan). This optimization eventually leads to cost efficiency for pay-per-use cloud resources. RDAS+ consists of partitioning, resource allocation and task scheduling steps to realize such optimization. We have evaluated RDAS+ using five types of real-world scientific workflows in comparison with three existing algorithms. Our experimental results confirm our claims on achieving resource efficiency. In particular, the average rate of cost savings (32%) outweighs makespan increase (11%). Although these two performance metrics are incompatible, the trade-off RDAS+ optimizes shows significant benefit particularly in clouds.