Recently several distributed rendering systems have been developed that exploit a cluster of commodity computers by connecting host graphics cards over a fast network to form a compositing pipeline. This paper introduces an algorithm for one such system to parallelize contributions from arbitrary numbers of OpenGL lights, for example to compute approximate soft shadows by sampling area light sources. The algorithm renders multiple shadow maps across nodes of the cluster by factoring the OpenGL lighting equation into illumination and material properties. Additional parallelism is achieved by utilizing multiple texture units within each graphics card. Illumination by L point sources can be rendered by Γ°L=KΓ ΓΎ 1 nodes where K is the number of texture units on each graphics card. For walkthrough applications each node requires a single rendering pass, while for scenes with dynamic lights or geometry K ΓΎ 1 passes are needed per node. We present results using fragment shaders on nVidia Ge-Force4 Ti4600 and ATI Radeon 8500 graphics cards, and a cycle-accurate simulation of the compositing operators. These results show interactive frame rates for walkthroughs and dynamic models rendering shadows with 32 light sources on 9 nodes of a Sepia-2a distributed rendering cluster.