I would be suspicious of a controller or even a sysex msg generated by the arpeggiator.  If you are sure you don't need any sysex for anything, delete all the banks.  (Of course, if they don't have data, never mind.)
 
Maybe you have already done this, but I would look at the event list and be sure all controllers are displayed.  For example, there could be a pitch wheel msg or a controller specific to this synth.
 
Another thing that can happen: another track on the same MIDI channel has a controller that affects your bass track during playback.  It doesn't mess it up when you click the note because when you stop Sonar, all controllers are reset. 
 
Have you tried soloing the track and turning off the MIDI input to eliminate the chance of a rogue controller in another track?

Well thanks for the suggestions everyone. The good news is that the problem seems to have magically disappeared tonight, although I sure as hell didn't (consciously) do anything to fix it. Well I never....this had me stumped yesterday, and it seems all it needed was a good old dose of 24 hours 

Coincidentally, I just experienced the same thing last night: one note intermittently playing an octave higher than written. It wasn't Trillian, though; it was a Kontakt library.
 
It was particularly frustrating because I wanted to freeze the track but had no way of knowing if the note would be interpreted correctly until I'd played back the frozen track. I knew there was nothing in the MIDI data except note and pitchwheel events, because I'd hand-planted them via the PRV. This was not X2, but 8.5.
 
I tried it again this morning, and the part played correctly. The computer had not been rebooted, nor had any processes been stopped or started, other than SONAR itself. Only the time of day had changed.
 
My theory at this point: sunspots. 

bitflipper
Coincidentally, I just experienced the same thing last night: one note intermittently playing an octave higher than written. It wasn't Trillian, though; it was a Kontakt library. It was particularly frustrating because I wanted to freeze the track but had no way of knowing if the note would be interpreted correctly until I'd played back the frozen track. I knew there was nothing in the MIDI data except note and pitchwheel events, because I'd hand-planted them via the PRV. This was not X2, but 8.5. I tried it again this morning, and the part played correctly. The computer had not been rebooted, nor had any processes been stopped or started, other than SONAR itself. Only the time of day had changed. My theory at this point: sunspots. 

I'm actually beginning to think that Sonar has a "time of the month."

bitflipperI tried it again this morning, and the part played correctly. The computer had not been rebooted, nor had any processes been stopped or started, other than SONAR itself. Only the time of day had changed.
 
My theory at this point: sunspots. 

My theory - it's simulating a real bass player - as you probably already know we sometimes have our own ideas about what "playing the part correctly" means. 

Oh great. Now, the damn thing has started playing the first few notes of a clip a semitone higher. I'm clearly going to have to rework this tune into free form jazz. 

Bitflipper, was it a bass note? IOW was it in the same range as Sharke's note?
 
And Sharke, is the notes that are playing a semitone higher in the same range as the bass note was?
 
Two different systems, two different programs, two different plugs/sample library, only one common thread here, sunspots!

With soft synths, the problem for me has almost always been other tracks on the same MIDI channel introducing rogue controller or pitch wheel messages.  I always turn off all inputs on a track after I am through creating it.  I also always have a Controller 121 = 0 and a pitch wheel = 0 at the start of a MIDI track.  Finally, I always channelize when I am done recording: use Interpolate to set all msgs on a track to the same channel so I don't affect other tracks and so all events are recgonized.  (This can get messed up when recording parts from keyboards not necessarily transmitting the right channel.) 
 
With external devices (not your situation), my problems have usually been sysex (same device numbers).  If you ever load a MIDI file from someone else, you can get some bizarre sysex stuff.