Back in the 1990s we used an almost identical circuit as yours, and eventually we found it was staircase saturation.
Unfortunately the only way to know for sure is to measure the drain current of both transistors simultaneously.
This means two identical and properly calibrated current probes.
The current waveform should be trapezoidal, but in the transistor which is suffering saturation the ramp segment instead of being a straight line becomes an exponential.
Simplistic contollers like the 3525, which was designed perhaps 40 years ago, does not incorporate a mechanism to prevent cycle by cycle volt-second unbalance, which leads to staircase saturation.
Later the Unitrode semiconductor company did bring out such a specialized push pull controller. But either the power supply engineers had become tired of push pull failures, or Texas Instruments which bought Unitrode did not promote the IC, but it never acquired widespread acceptance.
ADDITIONAL INFO:
Since you are using a variable duty wave form, you need to replace f with 1/Ton.
And most important, for N you use half the number of total primary turns, because that is how much you use each half cycle.