Hi,
for L2, all packets are processed in HW: unicast or not. Eventually, some may be flooded (unknown unicast), but still in HW.
for L3, packets may be slow-path forwarded if you have some HW table full situation, or if the packets are specific (IP options set, for example).
Master CPU will handle management and protocols. That's what goes to the Master. Anything else goes directly to the destination (that can be multiple depending on the packet type). The unit member receiving a packet does the lookup, find the destination port and send the packet directly to it. If it's on another unit member, the fastest path to it is used (it depends on the # hops and each stacking link speed - as you may have different stacking speed in your stack). Of course Master CPU will have the burden to generates ARP (if L3), and things like that.
Your stack is using a ring, and both links on each member are forwarding for unicast traffic. One link on one unit is blocked for broadcast, and one link is blocked per multicast groups. EXOS only output what port is blocked for broadcast.
If a stacking link fails, or a unit fails, failover happens very quickly and is impacting only the traffic that was flowing on that link at that time.
