OOPIFs have already launched to stable in M56.  They're used for web iframes inside extension pages and vice versa, and we have an active field trial for also using them to implement GuestViews.

I'm not sure if those use cases can affect headless, but there are more uses coming very soon.  We may need to prioritize this.  Can someone from headless take a look?

Can you point out where we assume there's only one renderer?

As noted in https://codereview.chromium.org/2830753004/diff/100001/headless/lib/browser/headless_web_contents_impl.h#newcode132, HeadlessWebContentsImpl::render_process_host_ is used in a way that assumes a WebContents has only one process.  That CL is also introducing more code making the same assumption (e.g., the test).

Not sure if there are additional places.

FYI, this is cl will make headless_web_contents_impl be associated with the main render process host: https://chromium-review.googlesource.com/c/chromium/src/+/673124