My personal take on it is that usually for stems the channel denomination should equal your other stems, or at least be LCR or 5.0. In the case of working in a stereo environment, I'd recommend going with stereo.
For one, dialogue does have to be panned sometimes - whether for creative purposes, or in rare cases to fix a story issue that's hard to see visual because of how its edited and shot (such as panning someone toward the left as they run out the building while panning someone in from the right as they're chasing them). To do this you'll want your DX sidechain to be multi-channel the whole way through even though 99% of the time it's going to sit hard center. All it takes is that 1%-of-the-time situation where you need to pan something and you'll end up kicking yourself because it won't let you pan or run a multi-channel plugin since you set your whole signal path up as mono (e.g. if you patch an audio track or Aux, in this case our mono DX sidechain, into an Aux stem submaster and/or into a print track for Input monitoring and THOSE are mono, PT WILL NOT let you pan anywhere else within the signal chain leading up to that final print track). It's confusing to try and write out, but basically you'll have your shoelaces tied to each other even before you've started the race, and only realize the problem well along the way once you've already written automation and made mix decisions - unlacing all that can be a nightmare, if not costly.
Also, Stems are what play back at unity gain as-is to create your Printmaster, so everything should be printed into your stems in final form already, including reverbs and delays and such. So that another reason to use multi-channel DX stems, otherwise all of your verbs are mono.
Thirdly, I don't like dealing with the "if it's in phase, the energy is +3dB/-3dB" math. With stems, since those feed 1:1 to the Printmaster and also serve as the archival media 'snapshot' of the final mix, I want whatever is printed in my stems to be EXACTLY how its going to read on the meters, dB per dB - not having some mono track where it relies upon PT to figure out on the fly that each channel of the stereo monitoring bus with have the dialogue distributed between each channel@ a -3dB attenuation of the mono track. Doing that means there's a 3dB discrepancy you have to keep in your head, especially if those Stems have to used for something down the road (such as working in a higher-order number of channels and doing a fold down, for example). If all the stem channel amounts are matching with none of this "on the fly" math such as dealing with a phantom center, you don't need to guess what the meters are saying and work backwards and hope something isn't clipping when maybe it is etc. What you see on the meters, for each channel, of each stem, is EXACTLY the sound energy which exists, hard-written, to each of those channels, dB per dB. Personally, I like that piece of mind knowing that everything is accounted for :)
