In this work we tackle the following problem: given a timed automaton, and a target set F of configurations, restrict its transition relation in a systematic way so that from every state, the remaining behaviors reach F as soon as possible. This consists in extending the controller synthesis problem for timed automata, solved in [MPS95,AMPS98], to deal with quantitative properties of behaviors. The problem is formulated using the notion of a timed game automaton, and an optimal strategy is constructed as a fixed-point of an operator on the space of value functions defined on state-clock configurations.