The exactly same issue happened to me earlier today while I was doing some split editing during playback. Suddenly all audio was muted but one bus output was at max level. Resetting audio engine did not heal the problem but saving the project and restaring SONAR again helped.
 
In the plug-in chain there were  Auto-Tune EVO, Waves CLA Vocal, Waves L2, PC-76 and QuadEQ. Bypassing/removing plug-ins did not restore audio playback.
 
SONAR X3e 64 bit, 44.1 kHz, RME UFX.

[EDIT: I was typing this reply at the same time that panup was typing his post above, so what follows may have been negated by panup's report.]
 
You're right, we can be 99.9% sure it's a software problem. I think we can eliminate SONAR itself as the culprit, since nobody else has experienced the exact same symptoms. 
 
That leaves third-party components and external dependencies (plugins, drivers, C++ runtime libraries and Windows libraries) and possibly even unrelated programs or services.
 
The big question: when it happens, does it go away when you globally bypass effects?
 
 
If not:
- Try turning off the "share drivers with other programs" option.
- When it happens, pull up Task Manager and note any processes other than SONAR showing > 0% CPU usage.
- Think about any software you've added since you last remember this problem not being there, and not necessarily just audio-related apps. Anything that could have replaced MSVC DLLs. It might be worthwhile to download the latest C++ redist package from Microsoft and re-install it.

>The big question: when it happens, does it go away when you globally bypass effects?
 
I didn't bypass effects globally; I deleted/bypassed them one by one (QuadEQ bypassed only because you can't remove it).
 
Somebody should confirm what happens with the global bypass in this case.

As an experiment disable intel speedstep.

Probably not related but since ozone was mentioned, I had a similar issue when using 32-bit ozone 4 in sonar 8.5, picking a random preset from the menu in ozone would stop the issue and then I could resume working, In this case the meter would peg at max and white noise would output from ozone during playback.

I've had this with ozone too.

If this was all hardware I'd say you're getting a feedback loop that feeds on itself. I wish I could give a more educated answer, but is there any chance that one of the signal processors could be doing just that? Routing an out to an in?
 
I'm really curious how you can get direct current on your outputs. If it really is DC then you will melt the voice coils in your monitors.

If I read the OP's post correctly he has an asymmetrical wave with only clipped negative values. There are many instruments including the human voice that record as asymmetrical waves but not to this degree. I would suspect a plugin or sonar's interaction with a plugin as the culprit.

Hey Bit - yes it does go away when globally bypassing all FX. Project plays fine then. The problem returns when FX are enabled. I've systematically removed different FX and find no obvious culprit. I even had a project with no FX at all (see original post) that had the problem, but that project was stripped down from a larger one with FX. Craig and Panup reported problems that potentially involved quite different plugins to the ones I use. I'm beginning to suspect this is an FX bin/chain problem, possibly VST3 related also. I think the FX bin is spitting out -Inf for some reason and under some conditions. This causes SPAN to choke, and Ozone to limit to the maximum negative 16 bit integer when exporting audio.
 
Cheers, Peter.
 
p.s Tried disabling audio driver sharing without luck - still got the problem.

ampfixer
If this was all hardware I'd say you're getting a feedback loop that feeds on itself. I wish I could give a more educated answer, but is there any chance that one of the signal processors could be doing just that? Routing an out to an in?
 
I'm really curious how you can get direct current on your outputs. If it really is DC then you will melt the voice coils in your monitors.

 
Artistic licence - sorry. I mean digital DC, that is, the exported audio file consists solely of the maximum negative 16 bit integer. Hopefully my audio interface doesn't actually play this as an electrical DC signal. They should be smart enough to filter DC. In any event, I haven't blown up speakers or headphones yet :)
I did check for feedback loops. The problem happens even with the Master out set to "none". This is definitely a software problem.
 
Cheers, Peter.