@swamptooth thanks for the details - no luck after adjusting all those settings. I like the Babyface a lot, it's definitely a step up from the Firestudio Project.
After a bit more testing, to summarize my issue:
With the RME, I can hear the bleed during playback. Bleed is also present in mixes. With the Firestudio, I can't hear the bleed during playback, but still hear it after mixdown. I had been experimenting purely with output pan, but input pan doesn't seem to make a difference.
I entertained the idea that the crosstalk is introduced by the device I'm playing the exported tracks out of. I've tried with my PC soundcard, my Nexus 5, and a separate laptop - crosstalk happens across all three. Tried a variety of headphones and speakers, no change. But the main reason I think it's not the device later in the chain is that I'm able to hear the bleed during playback with the RME in SONAR.
Maybe a dumb question, but the mixdown process is purely algorithmic, correct? So no hardware should have any bearing on what happens during the export itself (other than the CPU). Just trying to figure out any further avenue to test/explore this issue.