If the sweep rate is too fast the filter will exhibit tilt.
Also when I went t 82PF I used ATC 100B 5% parts I had so the result was less random.
I also found that the terminating impedance was better and the filter ripple was better with a
differently constructed transformer (ft43-2402 2 hole cores). I would later design a 100 ohm
transformer that gave better results (2.7khz and flatter).
Keep in mind we have no data across more than a small sample of crystals and maybe only
one vendor. So variation is expected.