OK, I understand a bit better now.
The suggestion of changing the eluent system, which some people gave you straightaway, might solve the problem from a practical point of view, but I seem to gather you're more interested in explaining why this happens from a theoretical point of view.
I'm still convinced my hypothesis could be correct.
I'm not surprised NMR detects the difference between the various stereoisomers: NMR, in a way, gives you a 'picture' of the nuclei in your molecule, i.e. it goes very deep into the structure, and the local environment of the nuclei counts more than the whole molecule, in most cases.
But when you have your molecules in solution and you let them interact with silica and a solvent, it's the overall, average properties that count.
Imagine you have two diastereomers RR and RS. Their ethyl derivatives RR-Et and RS-Et are well separated as you said, because the Et group is small enough, and the intrinsic difference in properties between RR and RS prevails. Now you make derivatives with a large moiety such as menthol, that we can call M, so you get RR-M and RS-M.
My idea is that M is so big that it overpowers the RR/RS part, so that the overall properties of either RR-M and RS-M will be more similar to M alone, making them more similar rather than more different.
Now, I know it doesn't work like this in reality, but just as an example imagine that the overall properties are proportional to molecular weight, and say that the RR moiety weighs 50. Ethyl weighs 28, whereas menthol weighs 155. So, the ethyl derivatives will weigh 78, the menthol derivatives will weigh 205.
In the ethyl case, the % of weight of your derivative that is contributed by RR is 50/78 = 64%, whereas in the menthol case it will be only 50/205 = 24%.
When you use menthol on two enantiomers, the fact that you generate asymmetry where there was symmetry is probably more powerful than these considerations.
But I might be totally wrong. This is just my hypothesis.