Sarah Ambrose, good information, thanks for sharing. A couple of questions that came to mind when reading your response.
In discussing Option 2, you put in bold "Only do this if the dataset is small (< a few hundred features)." I understand why you are emphasizing that fact, but I also wonder what Esri is doing to make users aware they are not following that guidance when they try Option 2 using a huge dataset.
The organization I work for has thousands, some may argue tens of thousands, of GIS users with a wide range of knowledge and skills. You might have your bold text here, and the warning may even be in some documentation somewhere, but I can guarantee 90% plus of the users in my organization will not see such a warning in the documentation or here. What safeguards are going to be in place to at least notify the user they are doing something ill-advised if they try Option 2 using 50,000 records or 500,000 records?
Regarding Option 1, I see that becoming problematic, even in the short term. Typically if people want to use distributed GIS, they are working with fairly large datasets. If everyone working with large datasets needs to copy them to an AGS data store and publish a service, and the derivative products are also going to be stored and published the same way, storage management is going to be a nightmare for IT staffs on the backend. It won't take long for even 100 users working with large data sets and generating lots of derivative products to burn through hundreds of terabytes of storage in the data center hosting ArcGIS Enterprise.