The way I understand the stratum protocol and desired Pool total network I/O coming from your location, it's recommended to set it according to total GHs power available to you on-site.
In your case (4 GHs), I'd set it to 4 then.
At difficulty 4, I get a lower rate of accepted shares. Is this still advantageous in some way? Each share in cgminer displays either device 0 or device 1, its faster with difficulty 2.