Computationally it should not be too much of a problem (if done properly): computign paths over a larger distance might take a little longer, but should not be impossible. I don’t know how they simulate passengers, but once a destination is established (randomly?) and a path chosen (fastest?) the agent simply follows its path until destination. Calculating a path is therefore only a minimal part of the time. However, I do not think that for passengers it is a good idea… it replaces most of the incentive to have a well connected and efficient network: the farthest (and quickest) I can bring customers within 20 minutes the more money I do. For cargo, this could be a good idea, though.