Browser-based volunteering computing projects are mainly used to perform scientific computations in heterogeneous clusters at a low cost. As for every community-driven approach, saboteurs can try to cheat the system for various reasons. In this paper, we propose to study whether such solutions could improve their performance and resilience by using shadow banning instead of a classic ban scheme. To do so, we have built a framework simulating a real system and studied the impact of shadow banning in relation with task types, saboteur rates, and detection techniques such as majority, m-first and credibility-based voting. Results show that shadow banning is overall more resilient, reducing the number of errors of detection by more than 33.5% in average. It also improves the server-side performance in a significant manner for saboteur rates between 0 and 20%.