To be fair, since I use an open-source backup script, I need to take into account the time I drop the Zimbra services for backup. It's been a while since I stood there with a stopwatch to monitor it, and I haven't modified the script to log stop & start (I suppose I should), but last time I watched, I was actually down for approx 2 minutes (and a big piece of that was the time it takes for the mailbox services to start, actually).
Since there are 1440 minutes in a day, 1438/1440 = 99.86%. Of course, that's robbing 2 minutes in the middle of the night when nobody is checking my email, so downtime that's visible to the end user would be less. . .
But if we look at involuntary downtime--that which happens because something broke, then I'm talking only a couple incidents since I started on Zimbra--and those were failures of the Clam database. That is the only thing that has ever caused me to LOSE zimbra service.