Ah, Sorry mis-understanding. I can see the differences in the mids, I just thought we were talking specifically about around the cut-off area.
 
I have no idea why there would be any change there. Of course with a snapshot picture it's almost impossible to tell what is going on but if the EQ is causing it, I'd suggest buying a better EQ or use the Prochannel.  <---- That is a joke!!
 
On a more serious note have you tried a different EQ to see if that is the issue?
 

This is normal behavior for an EQ, even if most of the time people don't realize it.

Sanderxpander
This is normal behavior for an EQ, even if most of the time people don't realize it.

I assume that it is the comb filtering caused by the slight phase delay then?

My guess is that most of the time people don't get near 0 dBFS until they have a "full" mix with some bass instruments and so when they use a lo cut a substantial amount of energy is cut which offsets the phenomena so as to make the overall effect less dramatic than a 2dBFS increase. Of course, if you have the EQ in front of a limiter you may not notice it even when it is dramatic.

I tried with some other EQ. None had a 48dB slope. I imagine the steep slope is more effective at cutting the energy and so the result is more obvious. The other EQ didn't seem to do it.

When I create a full mix with everything in the mix I am still nowhere near 0 dB FS. The rms level of the full mix is still down at the reference level which may be -14 or -20 dB in my case. That is how the K system works. Tracks at the ref level, buses at the ref level, the final mix at the ref level.  (For me) anything other than that is just plain wrong.
 
Because I work this way the problems that have been raised here  (and many others)  never have any effect in reality. (they are not important!) That is why it is just a great concept.  It eliminates most problems before they arise.  The issue that Mike is talking about here is totally insignificant.  I am not saying they don't exist, but if you are clever you can avoid them alltogether.
 
It probably explains the reason I am seeing no level changes. I was doing all the tests at the chosen ref level I was using at the time.  I did find the test signal in Mike's case interesting (cymbals) something I would have not thought of.  But in reality cymbals are often very low in the mix while they still can be heard very clearly.  Also if I was using a HPF on cymbals I would probably be pushing the cutoff frequency a little higher as well.  In that situation levels are usually going down not up.

Ditto Jeff's post. None of my mixes get that close. Having said that I thought the original question was what did we think was causing the peaking.
 
I'm going for the phase delay.
 
Artefacts are easier to hear when boosted which is why it's always better to cut rather than boost when using EQ.

Excuse me if I'm misunderstanding because I don't use K-level metering, but in my understanding this becomes a problem even before hitting the main fader/meter, as long as your audio file is significantly hot and your EQ is pre-fader (which is basically always). Or do you always gain down when using K-level?

K metering goes in to the red at -12dBFS, -14dBFS, -20dBFS etc. depending on how scared you want to be.
 
An example; If you have your system calibrated at K14, so that -14dBFS RMS equates to 85dBSPL at your listening position, and you do not notice that your EQ settings are adding 2dBFS to the peak digital output you can feel safe that no one else will notice any digital full scale clipping.
 
As far as I can tell, It is a system that relieves people of the burden of having to remember that the -14dBFS marker on their meter is 14dB below the top.
 
 
 
 
 
 

I just mean if you have a hot audio file, your EQ can still clip regardless of any meter or fader level.

Yes, that's true.
 
 
 
I'm of the opinion that the unique character of an indefinite percussion instrument made it more susceptible to the increase in peak level, due to decreased interference, than typical musical content that has a tonal center, harmonies, and harmonics that represent a song key.
 
I suspect that the extremely close proximity of the component frequencies in the cymbal crash made it eligible for the phenomena to have a greater effect on the output than it would on material that has more space between the intervals.
 
 
Concerns about clipping etc. are, in my mind, broad based concerns that relate to all aspects of mixing and may elicit all sorts of ideas about floating point mix engines, "digital headroom", canned-soul emulations, etc. etc. etc.
 
 
 
For my part, I was simply fascinated by the prospect of having a 2dB peak show up out of "no where". I must be easily entertained. :-)