We've had some success entering the addresses of just two core nodes as the IP address "range" and not using seeds.
If you've a fast M.A. Network, try turning all the default discovery options down to minimum too.
Another possible solution would be to create multiple scheduled discovery jobs that can run at different times. Each job limited to only a few subnets each. The initial setup I am sure would be time consuming, but with the multiple jobs you can kick off each "area" at different times to also not tax the network resources too much. I would set HOPs to 0(zero) as well otherwise you might start finding those subnets that you have never seen before that are configured on the switches. These might be the generic 192.168 networks that are defaults sometimes.
Just a suggestion on how it might be easier to manage a few jobs then taxing the network too hard with SNMP retrying and requests all at once.
I have had the same issue in that I discover all the devices on a fairly small network, but looking at all the interfaces takes a long long time. These are fully populated 6509 and 6513 switches that have 48port ether cards in most of the slots. My discovery will hang/time out eventually trying to read "STRAY PORTS" have to almost limit my jobs to a single device. My Discovery time out is 400 minutes