Hello David
A switch is able to classify as well as mark traffic. Marking can be achieved using DSCP values in the IP header, but it can also be achieved by modifying the CoS values found in the 802.1Q VLAN tag.
So what would be the benefits of using CoS markings compared to DSCP? Well, for one thing, you’ve already answered the most important reason: DSCP values are applicable from end to end since they remain in the IP header for the duration of the transmission from source to destination, while CoS markings exist only wherever the VLAN tag exists, i.e. on trunks and on voice VLAN connections. So the CoS’s scope is very limited (localized) while that of DSCP is much broader.
The use of CoS and DSCP is not an “either-or” situation. Each one plays a different role, and as mentioned above, each has a different scope. Typically, CoS markings will only be applied on a switch for traffic traversing a trunk or a port configured with voice VLAN. So really, CoS values are only useful for assigning a priority that will ensure the correct management of traffic over a single specific trunk link (or voice VLAN connection) while DSCP is used at every hop.
I hope this has been helpful!
Laz