A digital subscriber line access multiplexer that multiplexes streaming packetised telephony and/or video-on-demand traffic is studied. To this end, the distribution of the amount of work in a discrete-time single-server queuing model with heterogeneous sources and general packet lengths is calculated. Obtaining actual data concerning the distribution of this quantity is notoriously hard. How the tail distribution can be extracted from the steady-state probability generating function through a multiple-pole series expansion is outlined.