In theory this should speed up tests running on Postgres as no WAL will be written, at the cost of completely hosing the test database if the instance crashes (which shouldn't be a problem with a test database).
When comparing builds with and without this patch applied though I had really inconsistent findings. Some builds ran much faster, but most ran much slower.
We already start postgres with the following CLI options which minimises WAL anyway, which likely renders the UNLOGGED change mostly redundant:
-c max_wal_senders=0
-c wal_level=minimal
-c min_wal_size=1024