Is FRBR a data model?

Not exactly. FRBR is too abstract and too generic to be a data model: no practical database could actually be developed exclusively on the basis of the entities, attributes, and relationships that are defined in FRBR. For instance, FRBR simply mentions that an instance of the expression entity has a "title" as one of its attributes. What is the real nature of this title? Is it a uniform title? Is it based on the uniform title for the relevant work, with additional elements? Is it some form of the "title proper" found on a document? Should it be controlled by an authority record? etc. In the process of developing an actual database, you would have to make a decision about the precise meaning of "title of an expression". FRBR simply says that in your database, you will typically have one or more strings that serve as a "title" for an instance of the Expression entity. The exact nature of a "title of an expression" is usually in the domain of cataloguing rules.

If FRBR is not a data model, what is it then?

Rather a conceptual reference model. One of the primary objectives in developing it was "to provide a (…) framework for relating the data that are recorded in bibliographic records to the needs of users of those records" (FRBR Final Report, sec. 2.1) and for assessing their actual relevance. FRBR may serve as a reference point for testing the validity and robustness of extant data models and data structures. It can therefore be used to improve extant formats as well as to provide guidance for the process of developing new formats. It also can be extremely valuable in helping design OPACs.

If FRBR is not a data model, does that mean that it cannot be  "implemented"?

Yes, it can be implemented, but it first has to be "translated" into a data model and a format.

Can that format be a MARC format?

Yes – some experiments demonstrate that it is possible to implement FRBR using a MARC format. But MARC formats are underdeveloped when it comes to actual linking devices, while FRBR highlights the notion of bibliographic relationships, which are best expressed as actual links. In particular, most MARC formats are limited in their mechanisms for expressing links between authority records. New tools such as XML seem to have potential in that regard.

Does that imply that MARC formats are to be replaced with XML-based formats?

Fundamentally, this will be a political and economic decision. MARC formats have been in place for almost 40 years. The cost of a radical shift from MARC to XML would be huge. Some advocate keeping MARC formats by just encapsulating them in XML tags, while others advocate a more radical change. Anyhow, some kind of change will definitely take place to take advantage of the evolving computing environment.

Will it prove necessary to create one record per entity (work – expression – manifestation – item)?

The question is still unanswered for the moment. Experiments are being made in both directions (single-level "flat" records vs. four-tiered records). Implementation assessment reports are needed in order to definitively answer this question. Besides, not every kind of document needs "full FRBR treatment". Many works are available in only one version and are issued only once in only one format: the FRBR tree structure would not prove very useful in such cases, and would result on the contrary in much redundancy. The whole range of documents from Hamlet to self-published holiday recollections represents a continuum from maximal helpfulness of FRBR to no helpfulness at all.