even 10.20 didn't really support multiple default routes as you can guess from the zero use count on one of them in the netstat output. if you need/want default route failover, you might consider enabling the router discovery protocol daemon. it will find and "talk to" the routers on your local subnet and decide which one to put into the routing table.
of, if the two routers on that wire support proxy ARP, you could configure your own local IP as the default route (with a metric of 0 rather than a metric of 1). failover would then happen based on arp cache timeout settings.
as for login delays, if it is only login delays, i doubt that duplex on the NIC would be an issue, but you can double check that the NIC and the switch match. late collisions if the NIC reports half-duplex are an indication of mismatch, fcs errors when the NIC reports full-duplex (lanadmin -g mibstats
)
it could be that your DNS servers aren't being reached quickly - do you have more than one DNS server in /etc/resolv.conf on the server? you could take a tusc trace pointed at inetd's pid, told to follow forks, be verbose, display pids and have a timestamp and then do a login and see if you see DNS delays - you would first see a call to socket() for a SOCK_DGRAM socket, then some bind/connect on that file descriptor involving port 53. ftp://ftp.cup.hp.com/dist/networking/tools/
or you could use tcpdump (HP Internet Express or www.tcpdump.org) pointed at your lan0 interface and given a filter expression of "port 53" ("dns" might work as a shortcut)
there is no rest for the wicked yet the virtuous have no pillows
