It just seems weird to me that what they sound like should have any effect on how they're cataloged. Why should that matter? Why do we want to put that additional burden on editors? Anyway, you've probably heard their stuff. Here's something from each of them:
Getting to Know You
I Am the Very Model of a Modern Major General
Thank Heaven for Little Girls
The Lady Is a Tramp
I think long-term collaborations like these are rare in the Classical world, especially before 1920. Wikipedia has two articles of interest: a list of opera librettists and a list of Classical collaborations. The first list does show a few instances where the same composer and librettist worked together multiple times, but from what I've read, even then it seems more like the composer commissioned the librettist and they worked separately, and not a co-credited songwriting team like what was more common in the 20th century. The second list only has three entries I could find that really fit the description of a long-term collaboration: François Rebel and François Francœur, Luigi and Federico Ricci, and Federico Chueca and Joaquín Valverde Durán.
I'm really split on how to best approach this. The main problem is that with some of these teams, they're so well-known as a unit that users are going to expect to treat them that way, and MB is really bad at that if they're split. If I search Discogs for albums by Rodgers & Hammerstein, I get plenty of results. But on MB, it's a nightmare: they don't appear as a team in an Artist search, the Overview page for either individual shows solo, joint, and any other collaborations all mashed together, and the filter only allows one unique AC -- it treats split releases, additional performer credits, different join phrases, etc. as separate filters. The works page doesn't have a filter, if you want to browse those you have to use a search with advanced syntax. So it's definitely not the most user friendly thing in the world.
And it applies to any well-known songwriting team, not just Theatre or Classical: George and Ira Gershwin, Leiber and Stoller, Goffin and King, etc. The Sherman Brothers wrote nearly everything together, but have no joint profile. We even have a separate artist page for Holland-Dozier-Holland (granted, because they recorded one EP as such), but a few works that use the group with the credits, and Discogs uses their joint profile for most of their writing credits.
On the other hand, trying to combine all those those ARs would be a pain.
I definitely don't like the idea of singling out the exceptions. If there's a pattern to this, we should include the pattern in the guideline instead, and cite those as examples rather than exceptions. Something like: "Non-performing collaborations should be credited as individuals whenever possible. Work (and Recording?) relationship credits for duos or groups should be credited to individual members whenever known, even if the album and track is credited to the joint artist." People tend to do this anyway -- give composer credits to the members even if the release just credits the band as writers, but it's actually not covered in the guideline. Should it be the group, the individuals, or both?
I'm not sure what to do about the name formatting, but I'd be inclined to allow it to be as credited on the release. I don't see the point of forcing the credit to be "Gilbert & Sullivan" if you can't actually search for Gilbert & Sullivan.
Last thing, the link for Hart is broken in the Theatre guideline.