For the sake of completeness, I retested the original Noctua P14 (3-pin) with the FanXpert2 Auto Tuning:
Noctua P14 3pin FanXpert 2.jpg
Running it off the CHA_FAN 2 the results are more closer to your one, even if I know they aren't trustful: take a look at the same fans on the same headers when controlled by SpeedFan
Noctua P14 & P14r SpeedFan.jpg
No automatic tuning, but on the other hand I prefer SpeedFan over any proprietary piece of software (whether it works).
Well, my not that much educated guess
is that, given the NF-P14 FLX stops itself around a 24% duty cycle, FanXpert didn't properly recorded that fact, because this value is intermediate between the FanXpert2 setting points of 20% and 30%, so that it probably picked the higher one as reference.
On the other hand, my guess wouldn't seem to apply to you, because you should have three setting points correctly recorded under the reported reference min.
At any rate, IMHO those issues seem to be due/related to sort of a bad interaction between the FanXpert Auto Tuning algorithm and the specific mobo header
(maybe you could try the tuning with the P14s hooked onto the CPU_FAN header, in order to see if anything changes).