depends on the server. I have a machine here that's going on almost 3 years without a reboot. I have other machines that were rebooted in the past month due to software updates etc.
We used to get 5 9's uptime when we were running ksplice on the servers, but we no longer use it for the kernel updates.
We usually run 4 9's (less than 5 minutes of downtime per month). Of course, some of the servers that we host are maintained by clients, and those machines may be running at 98% due to their own actions.
Since we have different level servers, maintained by both staff & clients, it's pretty much impossible to throw out a single figure, without including time frames and server usage types.
I've got a machine here that uses about 2GB of bandwidth per MONTH and it's been online for 3 years without a single blip. By the same token, we have machines used for online stores with high database & i/o activity that are running 2TB of bandwidth without issues for 3 months. Which one is the more amazing statistic?