The term "native XML database" (NXD) is deceiving in many ways. In fact many so-called NXDs aren't really standalone databases at all, and don't really store the XML in true native form (i.e. text). To get a better idea of what a NXD really is, let's take a look at the NXD definition offered by the XML:DB Initiative, of which the author is a participant.
A native XML database...
Defines a (logical) model for an XML document -- as opposed to the data in that document -- and stores and retrieves documents according to that model. At a minimum, the model must include elements, attributes, PCDATA, and document order. Examples of such models are the XPath data model, the XML Infoset, and the models implied by the DOM and the events in SAX 1.0.
Has an XML document as its fundamental unit of (logical) storage, just as a relational database has a row in a table as its fundamental unit of (logical) storage.
Is not required to have any particular underlying physical storage model. For example, it can be built on a relational, hierarchical, or object-oriented database, or use a proprietary storage format such as indexed, compressed files.
There's a lot that we can learn from this definition, but the three main points can be summarized succinctly.
The database is specialized for storing XML data and stores all components of the XML model intact.
Documents go in and documents come out.
A NXD may not actually be a standalone database at all.
As should be clear from this definition, NXDs don't really represent a new low-level database model, and aren't intended to replace existing databases. They're simply a tool intended to assist the developer by providing robust storage and manipulation of XML documents.
Native XML Database Features
While not all NXDs are exactly the same, there is enough similarity to provide a basic discussion of features. We won't discuss any particular products here; if you're interested in evaluating products you can find a comprehensive XML database products list maintained by Ronald Bourret. Most vendors have first generation products available now, with a few of the more mature vendors having second generation products available. The NXD model is still evolving and will continue to evolve for several years. There is significant variation between products so the following sections just provide a high level guide that applies for most current products.
XML Storage
NXDs store XML documents as a unit and will create a model that is closely aligned with XML or one of XML's related technologies like the Infoset or DOM. This model includes arbitrary levels of nesting and complexity, as well as complete support for mixed content and semi-structured data. This model is automatically mapped by the NXD into the underlying storage mechanism. The mapping used will insure that the XML specific model of the data is maintained. Once the data is stored you must continue to use the NXD tools if you expect to see a useful representation of the data. For instance, if you're using a NXD that sits on top of a relational database, accessing the data tables directly using SQL would not be as useful as you might expect. The reason for this is simply that the data you will see is the model of the XML document (i.e. elements and attributes) rather then the business entities that the data represents. The business entity model exists within the XML document's domain, not within the domain of the underlying data storage system. To work with the data, you work with it as XML.
If a developer is comfortable working with XML tools such as DOM, SAX, XPath and XSL-T then they will probably be comfortable working with a NXD. The database will abstract away all the details of how the XML is stored and leave the developer free to build applications using XML technologies.