It is well known that the selfish behavior of users in a network can be regulated through the imposition of the so-called optimal taxes on the network edges. Any traffic equilibrium reached by the selfish users who are conscious of both the travel latencies and the taxes will minimize the social cost, i.e., will minimize the total latency. Optimal taxes incur desirable behavior from the society point of view but they cause disutility to the network users since the users' total cost is in general increased [4]. Excessive disutility due to taxation may be undesirable from the societal perspective as well. In this work we examine the efficiency of taxation as a mechanism for achieving the desired goal of minimizing the social cost. We show that for large classes of latency functions the total disutility due to taxation that is caused to the users and/or the system is bounded with respect to the social optimum. In addition, we show that if the social cost takes into account both the total latency and the total taxation in the network, the coordination ratio for certain latency functions is better than the coordination ratio when taxation is not used.