I'm a big fan of ksplice - and it's been great, but I still do monthly reboots on servers in order to flush anything out of memory that may be hanging around and keep the system running optimal.
The problem I have with uptime reports is that they're hard to guage. We have 400 servers, and while we may have 300 with 100% uptime over the past 5 days, if a user rebooted their server for any reason (dedicated server), or their site was hacked - that really shouldn't fall under OUR uptime performance as it was the fault of the user, not us.
Outside monitoring services monitor a domain or IP number for either a HTTP response or Ping response (depending on service). If their apache crashes for any reason, that's a negative mark. If they didn't pay their bill and their server is disconnected, that again is the fault of the user, and not us as a hosting company.
It's very hard to show public information as to uptime as a simple reboot may be necessary to upgrade a kernel or PHP and that then shows 99.98% uptime, even though it was scheduled maintenance.
We're asked all the time how long our servers have been up, but with so many, you can't just say 100% or 99.9% as there's many caveats as to why something may have been offline.
Even here on hosting discussion. For the most part, the server has been online and working over the past 2 years - but there have been 2 and 5 minute outages each month in order to upgrade scripts and keep things operating correctly. Is the site considered online 100% - technically, no.