The strongest argument as far as I can see it is... the problem is you now lose control over all those things. It's a black box with effectively no knobs.

Anyways, read for yourself, Pavlo & Leis get into it in detail, and there's benchmarks:

https://db.cs.cmu.edu/papers/2022/cidr2022-p13-crotty.pdf

https://db.cs.cmu.edu/mmap-cidr2022/

What am I missing? The transactional safety problem (the bulk of the paper) is solved simply with a single writer. Which is where you want to be anyway for efficient batching throughput (and isolation).

The other concerns seem to imply there are no other programs running on the same machine as the database. The minute that's not true (is it ever true?). Then OS will do a better job (as seen with LMDB etc).

I think it's telling that the paper focuses on mongoDB not LMDB.