The paper attempts to clarify some issues concerning the use, and usefulness, of the packet-speech multiplexer. An accurate mathematical analysis is used to find the trade-off between delay and packet loss. The subjective effect of delay and packet loss is determined using data from our own, and published, conversation tests. It is found that the multiplexer will operate most efficiently if queueing delay is limited, with a large delay limit for heavy traffic conditions and a small delay limit for light traffic conditions. The improvement in TASI (time assignment speech interpolation) advantage gained through allowing delay is small, and it is concluded that the packet-speech multiplexer is unlikely to be used purely for its buffering facility. However, its ability to multiplex speech channels where the information rate is continually varying may be of significance for the next generation of speech coders.