The Map/Reduce worked quite well (and fast), until I hit this size limitation. So now, it's a matter of finding a way around this technical limitation, hopefully without an architectural revamp.