This largely depends on the scalability, and (perhaps even more importantly) the scalability of the platforms you plan for your CMS to run on.
For example, a lightweight (i.e. AWS t2.micro) server would handle several (not hundreds, but enough) low-volume sites very easily. However, a single high-volume site (think Wikipedia, etc) is obviously distributed over many MUCH more heavyweight servers (not the plural).
Where this becomes a problem, is in the ability of one site to scale up. Let's say someone's political blog (running on your new CMS) gets linked from CNN.. In about five minutes it goes from a few dozen hits a day to hundreds per second. Now, obviously the underlying platform scalability is dependant upon your platform (AWS, Azure, actually hosting it on a real life computer under your desk, etc) - but the ability of your CMS to leverage the resources it has in such a situation is VERY MUCH impacted by hosting multi-sites rather than having a dedicated instance.
There is also the human element to consider. Let's say your multi-site instance is hosting sites from different users... Now when Fred's politics blog goes viral, Aunty May's Delicious Biscuits Website also slows down (or even goes offline entirely). This is a lot more difficult to explain to poor old aunty May than if Fred was the holder of all of the virtual sites in the instance (i.e. Fred's politics blog's popularity made Fred's scrap-booking site go offline).
Obviously there are performance benefits to be gained from things like DB connection pooling, etc for multi-site systems (otherwise they wouldn't exist). But make sure sites can be promoted as they become more popular (and require more resources). A small site that starts out as a virtual entry in a multi-site system may one day become a behemoth (perhaps not as large as Wikipedia, but still) requiring multiple dedicated servers and databases.
Overall, I would try and take a good look at your expected use-cases/users/etc. How likely is it that there will be popular sites (you can assume unlikely, but never impossible), how likely is it one user will want to have multiple virtual sites on this CMS? For that number of users is it worth the extra development effort? Would most (or all) of those users easily know how/be willing to just use nginx to manage it if you didn't do multi-site?