Good point. Obviously you would have to restrict the applied g-forces, therefore the time to stop would be longer. If I'm not mistaken, this should only a linear relationship though, e.g. if you restrict the acceleration, the time to do 10x delta_v is 10x the time you need for delta_v. As long as the earthquake forecast is good enough, this should be manageable, e.g. you need 300 seconds instead of, say, 30. 5 minutes forecast seems realistic to me. Obviously you can never give 100% safety - if the epicentre of a big quake is right underneath your trainline you might be screwed; However that chance is really small and we already live with these kinds of risks today.