I believe there must be a heuristic that determines which IPs are returned to clients by the seeder.
The seeder may have a list of hundreds of available nodes, but it only returns a reduced subset when queried.
Where in the code is the determination made of which IPs to return in response to a DNS query (out of all nodes the seeder is tracking and marks available)?
Note: The author of bitcoin-seeder (sipa) has stated that as many IP's as fit in a UDP data packet are returned (So there is no max_addr_return variable or such similar setting)