Collaborative search, where the activities of multiple users are combined to satisfy their information need, is an effective tool to to handle complex search tasks. People search collaboratively in groups of varying sizes. Various collaborative search systems have been studied in previous work, but they only investigate a fixed group size. Therefore, the impact of group size on retrieval effectiveness in collaborative search is an open research question. We investigate the effect of group size on retrieval effectiveness in collaborative search in a crowdsourced study with a total of 305 participants, in groups varying in size from one to six. We use a web-based system for collaborative search in this study called SearchX. We extended SearchX with two features for algorithmic mediation, which aims to support users in division of labour and sharing knowledge with collaborators. We investigate three variants of our system with and without features for algorithmic mediation to investigate its effect on retrieval effectiveness for groups of varying sizes. Our results show that the group recall increases linearly with group size. In contrast to a previous simulation study by Joho et al. [20] we do not find diminishing returns in group recall with increasing group size, suggesting that larger groups may increase group recall further. We also find that the investigated algorithmic mediation features do not significantly affect retrieval effectiveness. We conclude that the simulation results do not translate to our study, and that future collaborative search systems should be designed while taking the effects of mediation features on real users into account.