Great job on the demo, Geoff. I still maintain the differences are due to the test or the test environment, not the flac encoding. But what exactly that might be beats the heck outa me.
The reason I'm confident that any differences are not an artifact of the encoding is that the flac encoder actually generates an md5 hash from the original file. That hash is stored in the flac file so that when it's subsequently decoded it can be compared against the original. They have to match. Any encoder that can't do that is a flawed implementation of the algorithm. Hashes can't lie. It's the same basic mechanism that assures your network packets arrive intact, guarantees your disk data isn't corrupt and that you've entered a correct password. If just one single bit is askew in a 100 MB file, that error will be revealed when the hashes don't match.
One thing you could do is verify that the two imported files are lined up perfectly. You'll need to find a clearly visible transient in the file - maybe record a clave click - and then zoom in to the individual sample level. I don't know for certain that SONAR can even do that reliably. Something like Adobe Audition should be able to do it if SONAR can't.
It's hard to see in your video, but it looks to me as though the difference waveform is a series of large, abrupt level changes. I'd like to see that zoomed in and scaled to full screen. Abrupt changes are
not what you see when you do this kind of test with an MP3 or AAC, where you expect easily audible differences to be found. In those cases, you can clearly hear elements
of the original file in the difference file. Not random noise. (If you have Ozone Advanced, it conveniently has this feature built in. Unfortunately, only for MP3 and AAC, not lossless codecs.)