Apparatus, and accompanying methods for use therein, for illustratively implementing a large (e.g. approximately 1 Terabit/second) packet switch (200) or a non-buffer based statistical multiplexor (1810), using a crossbar matrix network in which, first, the output ports of individual switching elements (e.g. 1340.sub.1,1, 1340.sub.2,1) are partitioned into various groups (e.g. 1110) in order to share routing paths (links) (e.g. 1115.sub.1, 1115.sub.2, . . . , 1115.sub.K) among the elements in any such group and, second, the outputs of each such group are themselves recursively partitioned into a succession of serially connected groups (e.g. 1140, 1160) that each provides a decreasing number of outputs until one such output is provided for each corresponding output port (278.sub.1, 278.sub.2, . . . , 278.sub.N) of the switch. Such a switch also utilizes channel grouping to improve overall performance and a crossbar switching fabric that internally distributes contention resolution and filtering functions among the individual switching elements themselves to reduce complexity, provide modularity, reduce growth limitations and relax synchronization requirements of the entire switch.