I'm guessing they use this alongside CSPRNGs. Would make sense given the theorem that states any random number XORed with even highly-ordered input maintains its entropy.
There are lots of people in the crypto world who have serious issues with XORing random sources together.
I haven't yet seen a good argument why it's a bad idea, and part of me thinks it might be a way to get more software using "rdrand" or other insecure sources unmodified.
I think the bad idea stigma stems from people XORing from the same source. That totally is a bad idea, but if two sources are wholly independent, the maximum entropy in the combined systems is maintained.
To the people that just say it's never a good idea and scoff at any reasoning I'd remind them about OTPs. They are a special case related to this principle of XORing two independent sources together where only one input is random and it is proven mathematically to work.
I would use plasma spheres (e.g. [0], but there are lots of them out there). A single plasma sphere generates a visual display that changes much faster and is much less predictable moment-to-moment than a lava lamp -- so you wouldn't need nearly so many of them -- and uses much less power into the bargain.
"Prediction of a random stream from a lava lamp model constructed from entropic data inferred from encrypted packets"