Virtual worlds are nowadays commonly used in interactive applications, like computer games and simulations. Typically, such worlds are populated by a large number of virtual characters. On one hand, these characters have global goals with respect to the environment and thus, they must be able to plan their paths toward their desired locations. On the other hand, they should try to avoid collisions with each other and with the environment. In this paper, we present a new technique to improve the global routes that characters prefer to follow by taking into account the crowd density. Consequently, in our simulations, characters can intelligently plan their paths around congested areas, favoring less dense regions. This leads to an efficient flow of characters, reducing at the same time the amount of energy and time consuming avoidance maneuvers that the characters have to perform. The technique is fast and can plan paths for thousands of characters in real-time.