If that's the case, your issue is that the filter affects the amplitude of frequencies well below the corner frequency.
On the left is the spectrum of white noise with no filter applied, while on the right is the spectrum of the same white noise with the filter applied that I took screenshots of in my previous post (i.e. 6dB/octave @ 20KHz, which is the one I believe you're applying):
Sorry the scales don't quite line up (Audacity does that automatically and it can't seem to be changed), but as you can see, the filter begins rolling off at around 4KHz and is (for example) nearly 2dB down at 10KHz. This should definitely be audible.
To do this test correctly, you need to apply a very steep linear phase filter that doesn't even begin to roll off until above the highest frequency you can hear.
To my knowledge this isn't possible in Audacity, although I'm not 100% across all its functions. Certainly doesn't seem to be from what I can tell. What you want to do might require something a bit more specialised I think.
(As a side note, that kind of minimum-phase filter will also be affecting the phase of the signal well below the corner frequency, but I doubt that's specifically the cause of the difference you're hearing.)