There was a problem with the long polling monitor earlier this morning which has been corrected.  I was also making a few tweaks to prepare for what will probably be the final hardware upgrade (solid state drive for the database) this Thursday.  However, the slight backend tweaks should not have produced stale results like what you're referring to.

If you're running a slower miner (CPU/low end GPU), the stale % likely spiked due to the long polling monitor problem.