The effective_cache_size and shared_buffers settings from the revsys link are good, but thats generally older guidance ( max_fsm_pages is gone now ) If r17 doesn't use fsync.... Well. Perhaps you could show a postgres benchmark with that off. That's "Eat my data" mode. I don't recommend that mode, but it is faster, and might be more apples-to-apples.

MAINTENANCE_WORK_MEM and WORK_MEM could also help. Maybe. Thats a big dataset.

On SSD, random_page_cost may need tuned closer to sequential_page_cost.