Yet another area where I'm confused: When is it ever appropriate to use MST as a spanning-tree mode? I'm having difficulty thinking of a use case where it would be beneficial. Is this purely available for legacy purposes?
It's actually the newer of the types. But it makes things a little simpler to manage than PVST. PVST+ gives you the ultimate in granular control, but that translates into the potential for more administrative work to get things done.
802.1D came first
PVST was Cisco's interjection
802.1Q trunking took a jab at Cisco (required CST)
PVST+ was Cisco's answer
MST came about with RSTP
Rapid-PVST was the combination of the "best of both worlds" from Cisco's side.
The biggest benefits are in large numbers of Vlans provisioning such as a service provider edge services or large enterprises.
MST can drastically reduce the number of STP instances to a matter of design choice. It gets you operational, provisioning, computational
and protocol overhead efficiency while allowing you to run other STP protocols (in the same network ) for finer topological control, if you needed
it on certain vlans.
Sudeep Khuraijam is fully right about the benefits concerning spanning-tree instances if you use MST. I give you one example from what I have seen in real life in the past.
Using an old Catalyst 4006 switch and supervisor II (CatOS), Cisco best practices for that platform state not to have more than 540 STP instances. Unfortunately that switch was running as a layer-2 aggregation switch connecting uplinks from 6 other Catalyst switches. 160 active VLANs have been configured on 802.1q trunks on every link. PVST+ was running. That made a total of 6x160 = 960 STP instances. The switch was running on an average of 85% CPU, peaking to 100% and cutting off management access when changes were made to the STP. The STP processes took almost all CPU ressources.
After migrating to a MST using 2 instances (instance 0 for non-active VLANs, instance 1 for active VLANs), CPU load went down to an average of 40%.