With no autovacuum_vacuum_cost_limit and autovacuum_work_mem set the same as maintenance_work_mem, I wouldn't expect any difference between the performance of manual vs auto vacuum.
This was flawed. -1 for autovacuum_vacuum_cost_limit doesn't mean unlimited, it means inherit the manual vacuum cost limit which seems likely to be the default 200 units. Now that you have the manual vacuum done, it might be prudent to set the autovacuum_vacuum_cost_delay to 1 or 2ms globally, and watch to see if it stays caught up.