I haven't evaluated Markdoc fully, but I did have some experience using both MDX 2 and markdown-it—the parser used by Markdoc. I can honestly say that MDX 2 (and remark for that matter) is a lot slower in parsing bigger markdown files. Anecdotally, I saw markdown-it outperforming remark by a factor of 20 in terms of generating the file's AST alone.
Markdoc with its added complexity may add some penalty to the parsing performance, but does it do so to a point where it gets considerably closer to MDX? I doubt that this is the case, but would love to see someone verify this.
The primary difference between MDX and Markdoc is that an MDX article is essentially imperative code (like a typical React function containing JSX), whereas a Markdoc article is purely declarative. That makes it easier to reason about what's being rendered, and makes things like static analysis much more straightforward.
With MDX, you can essentially write arbitrary JavaScript code in articles. We had a similar situation when we used ERB, and it resulted in markup which was easy to write but very difficult to read. Markdoc strikes a better balance between the two in part because it has better guardrails and stronger division between content and code. It is a tradeoff, though.
If I’m honest though, the one thing that can beat it for me is my guilty pleasure - pug in Svelte. If a better way to author web pages exists, I’ve never seen it!
I'm currently doing an architecture decision record about Markdown documentation, and will add Markdoc to the candidates. The leaders so far are MDX 2 with plugins for JSX-style work, and Svelte for a fully dynamic site.
I'm aware of the Markdoc page about "Why not MDX?" which explains that Markdoc is deliberately less capable than MDX. But the page doesn't show how to do typical needs (IMHO) such as loops or substitutions. And for simple writing, compare with standard markdown annotated with tags/templates using Liquid or Jinja or similar?