Desmond, well beyond 'casual help'! What can I say?
I think we are getting closer to the reason of the problem, or at least what triggers it. Indeed, the 'trouble' bits always occur when Flex cannot track the pitch. So, if a long note will go 'husky' in the middle, the 'pitiching rectangle' appears broken, with gap. This is trouble. Just looking at either side (well, not just looking at) will cause the afore mentioned sound corruption of the unpitched section.
In general terms my 'speculative' explanation king of makes a bit of sense. Of course, in the old days of shifting samples in my AKAI S1000 the a rise in pitch would have caused a shortening of the sample, for obviuos physics. Now, any modern algorithm must employ some kind of time stretching to compensate for this. If the note that is shifted occurs only on the 'musical' side of the note, the associated (attached) beginning would be considered still part of the musical unit causing either a clip (if compressed) or granulation (if stretched).
I am certain that a dedicated professional tool such as Melodyne takes this aspect into consideration. Cubase did not 'chop up' a long note for instance as Logic does (in mine and your example): it 'bridges' the gap 'thinking' it's part of a whole.
Last, as far as my samples are concerned I am relatively shocked to see that they are indeed 16 bits, a standard that I have not used on a DAWN since... 2000. 100% certain that all my recording are 24 bits (as you can select in Preferences), thus maybe it happens on the bounce stage (I thought I had mastered this arcane art! Remember my stereo/mono issue). I will apply the dynamic action of a small hammer upon my family jewels if this were the case: I have bounced 20 tracks each of my 9 album project to be mixed soon! Darn!
Ok, what can I say? Thanks again.