So for a bidirectional(but always inserted at the subscriber) article, across publications, even if it is guaranteed that a uniquely identified row exists at only one subscriber (other than the publisher of course), you think it still might cause performance bottleneck? This is a filtered article in both publications.
As an example, let's say this is a 'UserLog' table. only subscribers log and not the publishers. There are two different publications, but only one 'UserLog' table, so we got to include it in both publications. We filter subscriptions by userid, so all datasets are disjoint wrt userid.
Do you think this still is a performance bottleneck?