A few latecomer thoughts on this already thoughtful discussion--
The difference between series represented by series statements only and those provided with standard access is better understood in terms of local decisions about access than some inherent property of different series. I agree with Karen's comment that series statements should be purely descriptive and not beholden to access requirements. The use of an identifier for a string to simplify repetitive data entry for series statements makes some sense, but is it really an identifier if it's only for a string, not an entity?
The FRBR solution to the relation between work and series is hiding in the concept of the aggregate work. A given text may appear within a series-designated volume in one publication and without the series in another publication. For FRBR, the text is an expression of a work, but the two publications are also works--aggregate works combining the text's work with other components. So the series has a work-to-work relationship with the aggregate work which contains the text work, and not with the text work itself. The exception would be cases where the series relationship is really a part relationship involving the text work, as between a book and the trilogy it belongs to. In such cases, the text work could have a part relationship to a larger text work while the publication work where it appears might have a series relationship to a particular edition of the trilogy, which would not be a true relationship between the text work and the series work.
I prefer models which treat series entities as separate entities with a relationship to items in the series and make the series entity description the locus for some of the series data elements that Joe and Theo discuss. The issues for implementing that model are as much organizational as structural. The ability to create community-level series authorities is limited to a subset of NACO members in our community, as the ability to create community-level serial records is limited to CONSER. There would need to be a way to extend the ability to create such community-level work to a larger pool of contributors if creating series entity descriptions becomes part of cataloging expectations generally.
Lastly, regarding numbered series. The relationship between a publication work and a numbered series is really different from that for an unnumbered series. The latter relationship is fairly simple, while the former is better understood as not just a relationship but an alternate designation of the publication work. Numbered series might better be modelled on the coding of standard numbering systems, with the series identifier specifying the system within which the number has significance, as "isbn" specifies the system for a specific number string. The way series numbers have been treated as appendages to the series title may read easily, but it's a case of a tail that needs to wag its dog in terms of data.