With more large-scale scientific computing tasks being delivered to cloud computing platforms, cloud workflow sys-tems are designed for managing and arranging these complicated tasks. However, multi-tenant coexistence service mode of cloud computing brings serious security risks, which will threaten the normal execution of cloud workflows. To strengthen the security of cloud workflows, a mimic cloud computing task execution system for scientific workflows is proposed. The idea of mimic defense contains mainly three aspects: heterogeneity, redundancy, and dynamics. For heterogeneity, the diversities of physical servers, hypervisors, and operating systems are integrated to build a robust system framework. For redundancy, each sub-task of the workflow will be executed simultaneously by multiple executors. Considering efficiency and security, a delayed decision mechanism is proposed to check the results of task execution. For dynamics, a dynamic task scheduling mechanism is devised for switching workflow execution environment and shortening the life cycle of executors, which can confuse the adversaries and purify task executors. Experimental results show that the proposed system can effectively strengthen the security of cloud work-flow execution.