TBH it’s probably true of everything in the metadata file, it’s just that I’ve never heard of anyone setting platform-dependent values for anything other than requirements, so from a practical point of view it’s just something we don’t have to worry about.
I think @jwodder is likely correct as to why Requires-Dist is treated differently, though that may be just a stroke of good fortune since it would be fairly common for the Requires-Dist information in an sdist to be inaccurate for a given platform.
I think this is one of the options for banning this “dynamic” metadata (I’ll keep using this term until we come up with something better, I guess), but it’s not really going to prevent people from continuing to generate “broken” metadata in this way. People will open tickets in pip or whatever project saying, “Such and such project has the wrong dependencies according to X command”, and then you’ll close the ticket with, “X should be doing the right thing”, and maybe X will hear about it and complain, “How the hell was I supposed to know this?” I doubt it’ll move the needle on the status quo.
I think we can come up with a transition plan to move people away from “bad metadata” and on to “good metadata”, but I think maybe it’ll take a decent number of developer-hours and might have to encompass more than just the Requires-Dist part. Maybe we can say, “OK, we’ll drop support for the legacy system even before we get our act together and start moving people away from it, since the things we’re dropping support for are all new features blocked on this anyway”, but I think there are more than a few things out there in packaging especially where the old way is deprecated and the new way is not ready yet 
. It doesn’t help our reputation to add another one of those things.