Why not just set dbnosync option as documented in the slapd-mdb(5)/slapd-bdb(5)/slapd-hdb(5) man pages?
I tried that as well, and it made the import faster, but at least in my setup using eatmydata still was a *lot* faster.
Another option, which I didn't think about when building my test setup, would have been to put the data directory into a ram disk during the import and move it to disk afterwards.
Either way, each variant seem quite similiar with regard to what effort they take, so I'm quite happy with my setup.