After the upgrade a large portion of the modified configuration of a slave is sent to the master.
Depending on the DB performance of the master, this will take a very long time and might
result in an - more or less - endless hang of the Node watcher process on the slave.
And subsequent changes from the master will not show up on the slave.

We solved our problem (9 MB of configuration data) by creating an additional index to the
DB table node_cksums.

create index node_cksum_index_2 on node_cksum (nodeid, cksumtype);

See also http://www.zabbix.com/forum/showthread.php?t=12226

This index should be integrated into the next release (1.8).

This "bug" is minor, because not everyone uses distributed monitoring and it depends on the number
of host/items/triggers/templates involved. Feel free to increase the priority.

We were in the lucky situation that this problem was reproducible, i.e., it happens a few minutes after
start of the (upgraded) ZABBIX 1.6.6 server.