In many settings there exists a set of potential participants, but the set of participants who are actually active in the system, and in particular their number, is unknown. This topic has been first analyzed by Ashlagi, Monderer, and Tennenholtz [AMT] in the context of simple routing games, where the network consists of a set of parallel links, and the agents can not split their jobs among different paths. AMT used the model of pre-Bayesian games, and the concept of safetylevel equilibrium for the analysis of these games. In this paper we extend the work by AMT. We deal with splitable routing games, where each player can split his job among paths in a given network. In this context we generalize the analysis to all two-node networks, in which paths may intersect in unrestricted manner. We characterize the relationships between the number of potential participants and the number of active participants under which ignorance is beneficial to each of the active participants.