There is no pre-set list of hosts distributed.

I assume the 5,000 hosts you're referring to are the cached hosts in your ~/.gtk-gnutella/ultras or other similar cache? In that case, let me reassure you: these are not systematically checked, they're just a pool for reconnecting to the network without having to contact a UDP Host Cache. That list is constantly refreshed.

But gtk-gnutella is not solely relying on this list of collected Gnutella hosts to bootstrap its connection: it also uses its saved DHT routing table, relying on the fact that all the peers listed there are necessarily modern servents (basically only LimeWires and gtk-gnutella hosts) which are, by definition, supporting UDP. Therefore, it uses them as a source of extra UHCs to probe for faster connecting... Since this probing requires only a single UDP Ping, this is extremely efficient, and the saved DHT nodes are usually stable enough (since they made it to the routing table).

The aim is to never rely on the central UHCs once a first connection to the network has been achieved. But in any case, random contacting of the cached hosts stops as soon as we get the desired amount of Gnutella connections.