I just saw this after an office move. The servers were offline for a while.
This is what I did for our CentOS Linux servers (similar, but not the same, steps would work for Ubuntu/Debian):
* verified/updated the /etc/ntp.conf file to be the same across the cluster
* stopped the ntp daemon: /etc/init.d/ntpd stop
* connected to the ntp server named in the ntp.conf file: ntpupdate <ntp_server>
* started the ntp daemon: /etc/init.d/ntpd start
* added ntpd to the startup manager: chkconfig add ntpd
* set ntpd to start in default reunlevels: chkconfig ntpd on
This cleared the offset issue. If this doesn't do it for you, you might also have to look into drift, or dig deeper into the time synchronization.