Thanks for reviewing.
Regarding the extra context, wouldn't that be the existing shared_context ? (code link)

// This worker should be accessed by threads that don't belong to the thread pool
// (in theory that's only the render backend thread so no contention expected either).
shared_context: Mutex<FontContext>,

It seems like async_for_each includes the shared_context in the list of contexts that it works on, so add_font etc. already do the right thing.

process_glyph calls font_contexts.lock_current_context:

pub fn lock_current_context(&self) -> MutexGuard<FontContext> {
    let id = self.current_worker_id();
    self.lock_context(id)
}

current_worker_id is a wrapper around Rayon::current_thread_index which returns None when not actually called from a worker thread, ie. when inlining.
And lock_context will return the shared_context in that case:

None => self.shared_context.lock().unwrap(),

... so TLDR I think everything is already working without contention?

But I'm not 100% sure and also I just tested with some casual browsing :) Please let me know what I've missed! Cheers.