No, not at the moment, though it's one of the last issues in our bugtracking system to check before we sign off the code and won't change any code before release. The thing is that it's a little hard to specify the relations, as the relations themselves don't contain the info about the field mapped onto the relation. So in-memory interpretation is a little cumbersome without that field. We're considering it a huge feature though, so it might be that we make the necessary architectural change and invest the few hours to implement this.
However it could be it's not doable because the predicates are interpreted at runtime, and do that themselves, i.e.: they don't use any outside info, they just get the entity and return true or false if they resolve to true on that entity. In selfservicing this could also lead to problems with lazy loading, which is also not that great (running a filter will fetch all data) but we'll see what we can do. For now, it's not supported, and if we find a way to do it we'll add it, otherwise it's for a future version.
