"What would be the advantages and disadvantages of this approach?"
Advantages (for this sized topology), little.
Disadvantages, amount of work to convert and additional complexity.
Other than you could do this, why do you believe it should be done?
I ask, because from the kind and size of the topology, seems unnecessary.
The critical "weakness" of OSPF, is running Dijkstra's algorithm, per area. What you describe, is up to 32 nodes, 30 with just a link to each core. Pretty trivial.
You also might be concerned by number of routes, like 300 plus in site 1, or about 400 plus, globally. Also that's pretty trivial.
However, an an alternative to using any kind of stub area, you can summarize an area's routes being advertised to area zero. (Of course for that to work well, area addresses need to be in nice summarizable address blocks.)
BTW, I've very large OSPF network experience, much, much (much) larger than you describe, and we didn't use stubs. Heck, I had one building using Catalyst 6513s with eleven 96 port line cards hosting VoIP phones and PCs. That one, port wise, equivalent to five and a half of your stacks, assuming stack members are 48 port. Such an OSPF node, was part of an area that encompassed other sites across multiple states, where sites might have multiple buildings and/or buildings of multiple floors and/or multiple L3 nodes per floor. Oh, and L3 OSPF platforms ranged from 800 series ISRs, switch stacks, to 6500s.
So from direct experience, again, why do you believe your network, as described, "needs" stubs?
