In doing so I have come to the following conclusions:
- For databases of non-trivial size and indices, mysqldump is not useful, because rebuilding the indices takes ridiculously long compared to any solutions that involve copying around raw data (replication, mysqlhotcopy, stop/cp/start).
- For replication over a WAN rather than a LAN, LOAD DATA FROM MASTER is not useful because it takes so long and so much bandwidth compared to locking, taking a local snapshot, unlocking, rsyncing to the slave, and resetting master and slave binlogs.