(I suspect that wasn't always the case. Especially over NFS or so?)
Still, if you have over 100k emails, mbox will be slow to parse for obvious reasons as it has to parse every line with often having mboxes as huge as 500MB if not more. Your FS and VFS layers will do a far better task on reading the first lines of thousands of files. In such case your mail client will parse the headers to generate the inbox layout in a much faster way.
No, you don't. You store in your metadata database what the offset of the message in the mbox file is. If you don't have such a metadata database, then you're going to be very slow no matter what the storage is.
Mbox it's just a literal
cat mail/*.msgs >> mbox
setup.
Reading will be slow, and file locking will be hell.
Unix should've switched to maildir long ago. Since 1999 storage and inodes are not a problem any more.
In other words, use mbox with an application that doesn't have its own message metadata database and you get problems. But if you're designing your own email client in the 21st century, then you're going to want to build your own message metadata database.
I mean, I studied Comp Sci at university in 1997 and we were given IMAP credentials