When you put the database server only 10 ms away (RTT wise) zabbix_server startup takes over an hour. When you put it on the same host it's only 1 minutes (server version 2.0.6)
From the logs it looked like it updated every single trigger from about 257011 within it's own update statement:
13716:20130929:080230.633 In DBget_trigger_update_sql() triggerid:604789 value:0(0) new_value:2
13716:20130929:080230.633 End of DBget_trigger_update_sql():SUCCEED
13716:20130929:080230.633 query [txnlev:1] [select distinct i.itemid,i.type,i.lastclock,i.delay,i.delay_flex,i.interfaceid from items i,functions f,triggers t where i.itemid=f.itemid and f.triggerid=t.triggerid and i.type not in (2,17) and t.triggerid=604790]

Which doesn't seem very efficient...