The proposed solution of an intermediate server caching embeds is needlessly complex. The first server a link is posted to can fetch the embed, then push it out to every server receiving the post.
Actually an interesting point. My immediate concern with that idea is that it would open the door for disguising things for what they aren't.
The solution was made from the general caution of not trusting remote servers regarding content they not necessarily control.
But yes, that would definitely be a solution, too.
Disinformation is a possible concern to be sure, but I don't think the solution to that needs to be technical. A server generating fake embeds seems like as good a reason for defederation as any.
If you’ve skimmed over the original publication, it is actually worse: it is a non problem. Other users have pointed out already: ~100MB served over 5 minutes period is quite literally nothing for even one small VPS serving the content independently, let alone a site with CloudFlare in front like they claim to have.