Maybe not the problem, but i've seen N75s behave differently depending on underhood temps, and sometimes other variations.
Re: hot days, load->boost conversion variance:
https://s4wiki.com/wiki/Tuning#IAT_effect_on_specified_load_and_requested_boostHowever, if you are running a lot of boost, and always want maximum performance, there is no point in increasing boost when it is hot, let alone reducing boost when it is cold. Also, as IATs rise, even with a perfectly flat KFTARX and LDIATA (see below), you will see more requested boost because the ECU knows that a higher P/R is required for a given cylinder fill (through the ftbr correction factor). To compensate, you may want to taper KFTARX across the board as IATs rise to keep your requested boost sane. Make sure LDIATA reflects those changes properly. Alternately, in theory, you should be able to find values for FWFTBRTA such that it cancels out the fixed IAT ftbr correction.
In particular, be aware that the maximum requested boost is slightly above the maximum MAP reading. Readers who understand PIDs should recognize that really bad things happen when the measured output cannot reach the setpoint.
If you have tuned your PID to really only work with a single set of req boost curves, if the req boost changes, it may not behave well.
This is a pretty big problem, and hard to solve w/o taking another look at your PID tuning approach; alternately, just try to make boost request look the same across a variety of IATs.
BTW you'll see the same problem with elevation variance.