Self-similarity and long-range dependence properties are detrimental to the network performance. Although these properties had been found existing in modern communication networks for more than 10 years, how to decrease their adverse effects on network performance remains a challenging problem to the researchers. One fundamental cause of the problem is: "what is the real statistical pattern of the heterogeneous traffic flows when they travel through the network, combine in some network nodes and separate in other network nodes". Previous work on statistical multiplexing under the self-similarity assumptions has led to many useful results. However, demultiplexing are also a very important part of the traffic modeling and analysis which is ignored by most of them. In this report, an in-depth study of the statistical multiplexing & demultiplexing loaded with heterogeneous ON/OFF sources is carried out, leading to the detailed and accurate results for the whole situation. we also find that the Queue of the MUX/DEMUX network introduces new dynamics for the demultiplexed traffics.