Use minimum-phase mode on tracks while mixing, low-latency linear-phase mode on the 2-bus when mastering. You won't notice any difference...at least, I don't. But then, I don't use linear-phase mode.
The benefit of the high-latency mode is supposedly better resolution at low frequencies, but in the mastering stage (which is when you'd use linear phase mode) it's rare (for me anyway) to employ precision high-Q filters in the low end of the spectrum, just broad cuts. The high-latency mode should only rarely be needed, if ever.
I'm not sure why Kontakt would weird out like that. Pro-Q's high-latency mode does allocate an unusually large buffer (24kb). This would be a good question to pose to the CW guys. I'd like to know the answer. Increasing your audio buffer size might help.
EDIT: while digging through the FabFilter forum I ran across this interesting comment by one of the FF guys, perhaps giving another reason for not using LP mode:
...low frequency edits in linear phase mode easily result in a "pre-ring" effect which can smudge transients