To answer a few of the above posts:
I'm not rendering any mp3's or wav's at this point, just recording guitar tracks for the project, directly into Sonar.
Yes. The clock source is stable. I have recorded many projects via the same setup, and this is the first time that I've had this issue.
When listening to the guitar part live, I do not notice this problem. I wouldn't call it phasing. More like just an occasional flutter. I didn't have time last night, but I will get some time tomorrow night, I can try a few things then.
I understand about the phasing that would be audible, if listening to both the live and processed signal through Sonar. My Roland interface has a blend knob, which allows me to sweep between the live input, and the processed Sonar audio. As you mentioned, I'm not hearing it during recording, only after the track has been recorded. It really does sound to me like the slight difference in quality that you hear, when you use audiosnap or melodyne on a track in offline render mode, as an example.
tlw
If you are both monitoring the audio directly theough your interface or from the Kemper and you have "input echo" turned on on the track then you'll be hearing the guitar twice. Once through the interface/Kemper then again when Sonar echoes it back to the interface some milliseconds later.
Depending on the latency setting the returning echo from Sonar might sound a bit like doubling, odd phasing which will reinforce some frequencies and weaken others or as a slap-back delay.
It shouldn't affect the recorded audio though, unless the echo from the track is then somehow being returned to the track that's recording so you are recording both the guitar signal from the Kemper and the delayed inout echo from Sonar. If that's happening then a phasing problem that sounds thin and a bit warbly might well be the result.
Might be worth checking your signal routing to make sure that isn't the case.