Usually, if a pair of SODIMMs seem to work perfectly fine when tested individually (including in both sockets), but failures are seen when these SODIMMs are used together, the issue is that the SODIMMs are actually different. I have seen cases where two SODIMMs, despite being marked identically and coming as a "matched pair", actually require (slightly) different timings. I would start by verifying whether or not the parts are indeed different. This can include inspection of the individual ICs used to see that they are identical, comparison of the SPD data, etc.
Hope this helps,
Well, a new development just came up - for the first time ever, I had a BSOD (CRITICAL_PROCESS_DIED) with only one SODIMM in the NUC. I rebooted into Memtest86 and it detected memory errors.
I've now swapped the other SODIMM into the same slot and it seems stable so far. I'll run the NUC over the weekend, and if nothing goes wrong by Monday I'll RMA the suspect memory module.
Thanks again for the helpful suggestions.