Earlier quoted context omitted.
The protocol notes that the relay has a choice to accept content, for example a large video. If you were running a relay you might want scanners for spam and illegal content running out front before relaying any content.
It's only recommended that you store json and text on relays. It's a text protocol, not a protocol for storing large binary data. Note sizes are restricted as well on most relays.
If I can store multiple messages on a relay I can distribute arbitrary files. So being text-only doesn't really limit anything.