From an architecture point of view, I'm convinced that the CMS is better treated as a service, and not as part of your web application. This is especially true in a larger organization where content needs to be collected and managed in different ways (via import, or where people edit or curate content).

At the start of this year, the company I work at had about six different CMS platforms. Many of the applications that used them were built on top of CMSs, and were tightly coupled to them. As for static content generators, we tried that. It really didn't scale, especially for dynamic content. Too many hacks, deploys, and hard to train people.

As of today, our team has switched nearly all public content for our company to LocomotiveCMS (http://www.locomotivecms.com/), and deployed a centralized multi-tenant system. In many cases, we use our Locomotive instances as an API, pushing content via the API or the command-line tool, wagon. The CMS then renders HTML templates (or even JSON), which are consumed by our applications via HTTP. Sometimes it's just a tiny part of a page, or sometimes entire mini-sites. As a result, most of our apps don't need to know about a database.

Whenever there's a change that requires a new model, or the addition of a field, it can be done quickly through the LCMS back-office UI (or the wagon CLI tool). We update the application(s) affected, and re-deploy. Non-techies can edit the content using the admin web UI in a familiar way.

In any case, it's been working really well for us. The CMS has become an API, a service, and a separate app, shared by many applications. We've become more flexible and efficient as a result, our web applications are no longer burdened by CMS frameworks or admin interfaces of their own. And we dont have to struggle with Sharepoint or Wordpress or any other nonsense.

It's been a fundamental shift in how we think about the CMS, and has scaled well across multiple projects.