After making a few simple changes in the database code, we're now syncing (no validation) at near full bandwidth capability on a 32GiB/10core machine (equivalent to Lopp's benchmark platform). Basically I was able to eliminate thrashing at any level by eliminating all significant reads while writing.