Sherlock Holmes needs clues

Doing a lot in acoustical engineering the first thing I would do is plot the frequency response in the room. You could do a simple 1/3 octave plot, but this is probably a little crude for what you are looking for--still it would be a start. I would actually use an MLS method, which is a noise burst and the an FFT transform that would give me the plot. The other thing it would do is give me reflections in the room and timing of those as well as the frequency range that they comprised. Then I would look at the waterfall plot and see what the energy time curves were like in this frequency range as opposed to areas outside this range. This would tell me if the problem was room interaction or not. Cases like this I usually expect it to be room interactions--but not knowing the room or equipment it's impossible to say. Even if I did know the room and equipment I would probably first do some theoretical calculations and then proceed with actual measurements.

Judth: We use the ETF software a calibrated microphone and pre-amp from Germany, and couple all of it through a PC computer. This provides the noise burst. The detector is at the listening position, but we generally vary it with about 6 sequential measurements to get some averaging (but do not actually use an averagin function as it can be misleading) It's pretty amazing the measurements we can achieve today with about 1/10 the cost of hardware required only 5 years ago. You can go to our website at see the software and measurement tools we use:
http://www.rivesaudio.com/software.html
The kits we sell here are primarily for our dealers that take measurements for us, but do still provide them to end users that feel they need this level of measurement for themselves.
The calculations are pretty basic if you are into room acoustics. Look at the impulse response and determine if it is being affected by the room and how much. If the impulse times match the distances for reflections, then it coincides with the room, if not, then you need to look elsewhere. The real issue becomes is there something different going on with the time energy curves (this can be viewed in the waterfall plots as well) in the frequency you are describing as a problem, vs the rest of the frequency spectrum. As Sherlock Holmes--I would look for these differences--in all areas of measurement (not just the ETC and waterfall) and then try to logically deduce why they exist and how they might effect the overall performance. I do agree with you--it is a Sherlock Holmes exercise, because you have to go after what you first suspect--it may lead you to nothing, in which case you have to look elsewhere--thus I have not provided a solution--just some areas to look into further.

Platos suggestion is a good one for testing the CD player. Whatever you do find, please do post it here--now I'm curious.