The main problem is this: The RAM makers always want to reduce their cost, so over time they may switch to ICs (memory chips) with higher density (less chips for the same capacity) for their RAM modules, or to cheaper and more available ICs which can manage exactly the same XMP speed/timings. So they keep the XMP profile exactly identical, to keep offering the same-spec product that people want to buy, but with different ICs under the heatspreader. This can then lead to the problems i mentioned above.
Namely, that it's not primarily about the speed and timings, it's about what kind of different ICs are used on the RAM modules, which need different parameters (electrical termination and so on, for good signal quality on the memory bus). But some parameters, they work for all the RAM, so it has to be a compromise. It usually ends up being a compromise that is worse than either of the two kits alone. Sometimes it might only boot after manual intervention, or it might not work at XMP speed. So a kit of two modules only is best, followed by four matched modules (in essence, a kit of four modules that are 100% identical and tested to work with each other).
To really see how identical the modules from the two kits are, read out the RAM module details with
Thaiphoon Burner. Run as admin, click on the "Read" icon and read the so-called SPD information. Then you can go to "File" - "Take a screenshot".
Also, something you can try which takes only five minutes: Swap the kits' places with each other. Put the first kit from A1+B1 into A2+B2 and vice versa. Sometimes the memory system like one combination better than the other.
