Spam is a problem, sure. But spam is somehow not objective and can touch the censorship grey area. I would prefer to control the spam filters on the client, so I can manage them and be absolutely sure about the rules.

Of course this approach has a cost: more cpu and bandwith usage; so the relays should mitigate abuse and DoS, implementing some throttling mechanism / smart firewalls / restricted accesses, to avoid to broadcast trash events. This is undoubtedly complex.