Perfect question! If anyone can provide such stats this would be amazing. :)
I would make the question a bit more specific. Let's assume I'm an author who wants to publish an article and have to choose between a particular toll-access journal and an Open Access one. We want to know what's the potential audience of both journals, measured by the number of scholars who can access the article, without differentiating between disciplines (for simplicity) and without considering non-scholars (they're hard to measure). The main question is how these two numbers relate to each other: the toll-access audience is 90%? 50%? 5%? ... of the OA audience? I suspect the value is closer to 5-10% for most journals, because there are something like 30,000 journals out there and most of them are not very presitigious and subscribed only by a fraction of institutions, esp. in developing countries. Being able to quote such a percentage value would make a powerful argument for OA.
So, one half of the answer is simple: potential audience for any OA journal is in the range of 7-9 million - that's an estimated global number of researchers, according to the STM Report 2015.
The other part of the question is more difficult: for a particular journal, say, Nature, what's the number of scholars who can access it through institutional subscriptions, and what fraction of 7-9M it is? Or, how many institutions subscribe to a given journal? -- I don't know the answers, but I think it should be possible to come up with some estimates by analysing subscription lists of different institutions and coupling these data with counts of researchers at each institution - if not globally than at least for a particular country. For US institutions SPARC may have some data, I guess?