If a war file contains several Persistence Archive jars, JBoss starts only the persitence unit(s) defined in the first jar it scans, ignoring the remaining ones.
For instance, given the following war file:

my-war
  +-- WEB-INF
     +-- lib
        +-- pu1.jar
           +-- META-INF
              + persistence.xml   // <persistence-unit name="pu1" >...</persistence-unit>
        +-- pu2.jar
           +-- META-INF
              + persistence.xml   // <persistence-unit name="pu2" >...</persistence-unit>}}

Jboss will only starts the pu1 persistence unit.

And if the war file has the following structure:

my-war
  +-- WEB-INF
     +-- classes
        +-- META-INF
           + persistence.xml     // <persistence-unit name="pu0" >...</persistence-unit>
     +-- lib
        +-- pu1.jar
           +-- META-INF
              + persistence.xml   // <persistence-unit name="pu1" >...</persistence-unit>
        +-- pu2.jar
           +-- META-INF
              + persistence.xml   // <persistence-unit name="pu2" >...</persistence-unit>

Jboss starts pu0, and skips the PU's defined in pu1.jar and pu2.jar.

This behaviour is non-compliant. Section 6.2 of the spec states that:

A persistence unit may be packaged within one or more jar files
contained within a WAR or EAR, as a set of classes within an
EJB-JAR file or in the WAR classes directory, or as a combination
of these as defined below.

[...]

It is not required that an EJB-JAR or WAR containing a persistence
unit be packaged in an EAR unless the persistence unit contains
persistence classes in addition to those contained in the EJB-JAR
or WAR.

And from section 7:

7.1.1 Responsibilities of the Container
At deployment time the container is responsible for scanning the
locations specified in Section 6.2 and discovering the persistence.xml
files and processing them.

When the container finds a persistence.xml file, it processes the
persistence unit definitions that it contains.