Here's the situation:
Today I was making a change to a peer on BGP. I changed the update source and pressed apply. At this point, I noticed the uptime counter counting backwards as the BGP process reconnected to it's peer. After it finished counting down to zero, it started counting back up. Maybe it was a de-sync between winbox and the host. Who knows? If mikrotik would like, I'd be happy to post a supout when it happens again. But it would be hard to capture, since it happens so fast.

Many months later, I'm seeing a similar bug in a different context - RB1200 running ROS 5.8 and 5.9 as a PPPoE server, and sometimes Winbox (running under Wine under Linux) shows PPPoE session uptimes count backwards to zero, then forward as normal. Disconnect such a client (click "Remove"), then it connects again with uptime typically set to "minus" (not displayed) a few minutes and counts down to zero. I've just upgraded to ROS 5.10 (due to stability fixes for the RB1200), waiting to see if it happens again...

Following up myself (moderator please append this to my previous post, not yet approved). It seems to be a winbox issue (just happened with 5.10 too), shows pppoe uptime counting down to zero then up, but the CLI shows the correct uptime (longer, as if counted from zero up). As I mentioned I'm running Winbox under Wine (version 1.3.15), under Linux (Ubuntu 11.04). No idea if it happens in real Windows.