My thoughts: the SSM2044 datasheet do suggest you should use the inverting input for a second oscillator, but you have to build a 3dB difference so that the oscs do not cancel completely when out-of-phase.
That's all fine, but I have checked all the schematics of synths using the SSM2044 and none of them use that inverting input, which is strange. I've looked at Waldorf, Siel/Opera, Korg Trident, etc...
Most use that to remove offset from the first input. I also more-or-less remember Olivier Gillet (of Mutable Instruments) writing about some issues with offset on the SSM2044 at the time he was designing that filter board for the Shruthi-1 (said filter board is since discontinued but I believe I saved schematics somewhere).
So we're faced with two things: the original SSM advice of using -IN for the second osc, and also no practical implementation as such, probably because of that offset issue.
So Chip, it's very worthwhile if you managed to test it. I actually was going to do the same with a single voice until I zapped my Arduino Uno!
In the meantime, I checked Johannes' Noise mod and he just mixes in with Osc 1 through a 33K resistor. If I'm not mistaken, in the Trident and some other synth with several 'oscs' before the filter stage, they do the same.
I guess both work, and the mixing one is very easy and we're 100% confident it should work since there are real-world implementations. The other solution of using -IN could be less than ideal despite the original recommendation because of that Offset issue.
I may test using the simple mixing solution, first with Noise, then with inputs to each filter. I should be getting some 33K resistors soon.