I can sympathize with the pain of a slow mbox parser, but even a magical parser can't resolve the fundamental flaws in the format.
I'm not sure if it's upstreamed yet, but at least the Red Hat fork includes a patch to use unlocked libc IO which more than doubles the mbox parser performance: https://github.com/karelzak/mutt-kz/commit/0672404fc710b4af5...
Mutt's parser is woeful, it's a decade or so of hacks, and in some cases it can't even round-trip its own mboxes without corruption
Having said that, i use mu4e these days, because i've become an Emacs user. With a few tweaks it now scratches my itch very effectively (except of course the HTML-email problem as mentioned elsewhere -- this is especially troublesome at work where they want me to include an HTML-embedded-image-tabled monstrosity with about a page of legalese as a signature).
That said there is an obvious solution to slow MBOX folders, which wouldbe to switch to using Maildir-storage, then using the `header_cache` setting(s).
Mutt offers quite good header/index caching, and there are other tools for improving on that.
A couple of years ago I wrote an mbox parser that was several times faster than Mutt's. It really showed because at the time my server was running on a Soekris net6501, my mailbox was a few hundred megabytes, and my habit is to invoke and exit mutt repeatedly rather than using a dedicated terminal window. Since then I've been running my server on a Xeon E3, and it's less bothersome.
I investigated patching Mutt to use the parser, but it seemed a little too much trouble for the effort given how many one-liners have accreted in the existing implementation. Using an index would make things more responsive, too, but the faster parser could provide competitive responsiveness into the gigabyte territory while keeping things simple, in both code and on the filesystem.
Is this a common enough gripe that it might be accepted and appreciated? My parser used Ragel but I could implement it again in plain C so it didn't have a build-time dependency.