Wissner-Gross and Freer has a 2010 paper in Physical Review on Relativistic Statistical Arbitrage [official version, pdf]. It makes the observation that as the propagation of prices and trades reach relativistic velocities, that very relativistic limitation of propagation induces some intermediate locations (midpoints between each pair of the 52 world exchanges weighted by turnover velocity) between exchanges from which profitable arbitrage can slow or even stop information propagation. The subject of the study is a Vasicek model of cointegrating instruments based an Ornstein-Uhlenbeck process and an Ehrenfest urn model of diffusion. One hypothesized strategy has to do with offsetting positions in geographically separated exchanges. Because proving one has the correct offsetting position in a distant exchange takes too much time, a low-latency system placed in the intermediate point between two exchanges can certify offsetting positions for both exchanges.
This does remind me of another article on the proposed London-Tokyo fiber link to run under the Arctic circle. The fiber infrastructure isn't always laid according to great circle distance. Many pairs of exchanges probably are not directly connected with each other at all. Moreover, turnover varies, perhaps this means one would need a mobile or at least a chain of load-balanced low-latency systems spanning the distance between each pair of exchanges to adapt to changing conditions. Perhaps there will have to be some dynamic hedging strategies too. Regardless, the paper is an interesting look into unintended consequences of the race to zero.