That's not supposed to happen... Maybe it's a filtering side effect. There are no "perfect filters" and some have bump before cut-off and some have ringing, etc. "Audiophiles" sometimes claim the sound is better if goes to 30 or 40kHz, but scientists disagree.
In fact the opposite is supposed to be true... You might be able to hear a loud-pure 20kHz in a hearing test, but our ear's sensitivity is lower at the highest frequencies, and the highest frequencies in music aren't "notes" but weak harmonics that are drown-out by slightly-lower frequencies.
And I don't think this is the explanation, but as you may know, the filter cut-off point is defined as the -3dB point no matter how sharp the filter. So a 5kHz low-pass filter is 3db down at 5kHz (as-is a 5khz high-pass filter). If it's a sharp filter that doesn't matter because it would pass 4990Hz with almost no attenuation.
Yes, I know all that, which is why I mention it - unexpected which makes it interesting. Yet repeatable.
Of course when doing the test we need a signal with strong HF energy. I use a high quality close-miced recording of jangling keys. Some music, like castanets or other small percussive sounds, can also work.
Also, when ABX testing subtle differences near thresholds of perception, we typically reach a zone where we can differentiate them better than random guessing, but we can't pinpoint or articulate exactly what the difference is.
PS: it's similar to listening to 2 identical clips, yet one is 0.2 dB louder than the other. We might detect the difference but we not as loudness, instead we might perceive it as sounding slightly "richer" or whatever. Just cuz we can hear a difference doesn't mean we can articulate what it is.