Not sure how much technical jargon you want to delve into regarding this subject, but this semester in my attitude and control systems class we had to do a project regarding autopilot performance.
http://www.soe.ucsc.edu/~elkaim/Docu...navigation.pdf I used that site as a primary reference. Very interesting stuff dealing with convergence and divergence of flight paths. That "5%" that you are refering to can refer to any number of configurations with the autopilot. Either the system can choose to not flare the aircraft leading to a hard or even destructive landing or in some cases the autopilot can go from a stable second order system like this
to an unstable system in which fishtailing and "unstable searching" can occur in the flight path. This would necessitate an immediate go-around, abandoning the approach and trying again if full scale deflection occured. Also in terms of vertical guidance the aircraft can exhibit the same behavior in terms of unstable pitch oscillations which could potentially lead into a stall as the aircraft is already operating in the back side of the power curve. Anyways, interesting reading in that link I gave you even though it mostly deals with GPS signals being used to verify position.