It is best to do planned system maintainance during off-peak hours. Adding or removing servers from the cluster will disrupt bin assignments and may interrupt service for some users.
Load is leveled based on the source port if available and the source address otherwise. Packets from a given address/port combination will be consistently routed to the same server by all levelers. Provided that the source information is sufficiently random, load will be equally divided among the servers. Bias in source addressing for a service will result in uneven leveling. Services with this characteristic should not be used with the TacoPleX system. It is the opinion of the author that services of this sort are rare.
TacoPleX logs messages using the syslogd facility. See its documentation to configure output. The messages report error conditions that may or may not cause an interruption in service. TacoPleX has a debugging mode that can be activated by a small change to the kernel source. It produces _lots_ of output and should never be used in a production system.
You should configure tacoplex.monitor to use a variety of alerts. Logs should be kept, but because any of the levelers could detect an error, there should be a centralized mechanism as well (or the logs could be spread out across machines). Email notification or paging seems the way to go. In the event of a network failure, though, these services may be unavailable, so logs should still be checked periodically.