Volunteer computing (VC) is a type of Internet based parallel computing paradigm, which allows any participants in the Internet to contribute their idle computing resources towards solving large problems. In VC, sabotage-tolerance mechanism is mandatory to improve the reliability of computations because some malicious participants may sabotage their jobs by returning incorrect results. Then, this paper evaluates current sabotage-tolerance methods such as voting and presents an impact of colluding attack on the sabotage-tolerance performance. The colluding attack is one of the most important issues to realize reliable modern VC systems because it may ruin the fundamental assumption of voting, i.e., "the majority is correct". This becomes a big threat to the voting-based sabotage-tolerant methods which eliminate malicious participants based on whether their results are the majority or not. In this paper, we propose a colluding attack model, and conduct a detailed study for two major voting-based methods (i.e., M-first voting and check-by-voting) through a Monte Carlo simulation to reveal the sabotage-tolerance performance against colluding attacks. Our model introduces a colluding probability to cover various sabotaging and colluding scenarios. The simulation study shows that check-by-voting works well even if colluding attacks happen, while M-first voting does not.