Hello Mark,
Routing any given virtual pipe simultaneously to more than one output perspective (by turning up the relevant 'perspective mix' on the voicing screen, etc.) uses no more polyphony than routing it to just one perspective.
The audio engine can process all four output perspectives simultaneously in each CPU instruction (via the AVX instruction sets, which allow 8 floating point numbers to be operated on simultaneously). (On older CPU that didn't have AVX, or on Apple's new Apple-CPU Macs which only have the ability to process four floating point numbers simultaneously in any given CPU instruction, using output perspectives 3 and/or 4 will be more CPU intensive, but it still won't count towards the polyphony limit.)