Ray tracing large-scale 3D scenes at interactive frame rates is a challenging problem on mobile devices. In this paper, we present a mobile ray tracing system that aims to render large scenes with many millions of triangles at interactive speeds on a small-scale mobile cluster. To mitigate performance degradation due to excessive data communication on mobile and wireless networks with still high latency, we employ a tile-based rendering strategy where each participating mobile device keeps an entire copy of the necessary rendering data. To realize such a system, we compress the 3D scene data to a size loadable into graphics memory, which enables an effective mobile GPU ray tracing. Also, by using a careful interaction scheme between the master and slave devices in the mobile cluster, we enhance the efficiency of the mobile distributed GPU ray tracing markedly.