Okay, I got it, but if memory requirements become too large you may still want to opt for a hybrid approach, parallelize over LOO sets at a coarse scale and internally over placements within each LOO task, like this large memory requirements can better be handled.