My advice is going to be limited, then; I can only tell you how I've observed A-B controllers on the network.
The connection failure time is negotiated at the time of the connection; the timeout multiplier is transmitted along with the other aspects of the connection like the connection ID. In every implementation that I've seen of EtherNet/IP, the connection multiplier was 0, which means "default 4x API".
The controller will consider the I/O module connection timed out after 4xAPI, and likewise the I/O module will consider the connection timed out after 4xAPI.
The connection originator (the controller and its originating network bridge) will send a CIP Connection Close request after the timeout but of course in the case of an actual disconnection there is no reply. If for some reason there is a reply, the connection is placed in a shut-down state by both devices.