There are a number of goals in doing this. There are a couple of approaches:
The algorithm is hard to explain. Basically, a penalty number will be applied to a route that flaps based on the number of times it flaps and how often it flaps. This number will decay the longer a route is up. There are a number of configuration parameters and some implementation parameters.
When a peer come up and there is nothing in the history record, everything is initialized and no overhead is in use. When a route becomes unreachable, some overhead will result.
Implementation in gated is pending. Both the space and time overhead are small.
What about memory requirements? For 64Mb routers, it will work fine.
What happens if folks set their paramenters differently? Nothing. It will modify whether certain information learned by EBGP is trusted, much the way a manual policy change could suppress a route that was flapping until it was OK to use it again.