Excess is an allowance for when peak shaping (or policing) is configured rather than average shaping (or policing).
Basically, during a time interval (often much less than 1 second, often in the range 25 to 40 ms), this number of bits are "allowed" (much as the sustain value does too). (Of course, when you exceed limits, excess is either queued [shaping] or dropped [policing].)
As I recall, some of Cisco's documentation does (or did) explain excess allowing to "borrow" extra bandwidth capacity, allowing a short "burst" of traffic, but still providing an overall average limit.
However, the common Cisco implementation, I believe, doesn't "borrow" but just adds/allows for a second allocation. So, for example, for your posted target rate of 128 Kbps, average shaping/policing would only allow a sustained rate of 128 Kbps. But peak shaping/policing would allow a sustained rate of 256 Kbps. (The latter, in such a implementation would, I believe, obtain the same effect as if the average was doubled. I.e. 256 Kbps average shape/police same as 128 Kbps peak shape/police.)
