*large meaning literally billions of users/devices in a group. Lots of other stuff will break before MLS
I heard that even MLS struggles with anything over 100,000K users to keep keys in sync. At some point the overhead of the protocol breaks it's own hability to send messages.
I've heard on of the authors say billions... Removing a member of a group takes log n binary tree operations, so it really does scale exceptionally well in this regard.
Def there is no better option