Well, when I define a "voice", what I'm referring to is the "source + modifier" part of a chain. So if you look at the Ts-L/Plaits pair as a "source" thanks to the 2-in VCA/mixer and the Sinc as the "modifier", that's how that should work. Remember: one VCO is good, but TWO of them means you can do quite a bit with detuning, waveform combination, etc + some form of mixing to sum those signals. And likewise, the Piston Honda has that dual oscillator architecture, then it feeds right to the Morgasmatron...but in this case, you can keep the entire "voice" in stereo, where the other "voice" is mono. And at the same time, you could just as easily patch those directly to the ES-9's inputs to stripe those directly to the DAW.
But the Cosmix is still pretty essential. Let's say that you come up with this complex result that requires the Cosmix's AUX send/return architecture to work with the Morphagene. Well, without the Cosmix (or something similar), you lose the ability to parallel process that way. I put it in because there'll be times that the working method from above is just fine, but there WILL be other points in time where you'll want to patch up something epic...and the stereo mixer/AUX architecture makes that much easier.