In EBCDIC:

x1 = '<><.A.B><. . ><.E><.F><>'

PARSE VAR x1 w1
           w1   ->   '<><.A.B><. . ><.E><.F><>'

PARSE VAR x1 1 w1
           w1   ->   '<><.A.B><. . ><.E><.F><>'

PARSE VAR x1 w1 .
           w1   ->   '<.A.B>'

The leading and trailing SO and SI are unnecessary for word parsing and, thus, they are stripped off. However, one pair is still needed for a valid mixed DBCS string to be returned.

PARSE VAR x1 . w2
           w2   ->   '<. ><.E><.F><>'

Here the first blank delimited the word and the SO is added to the string to ensure the DBCS blank and the valid mixed string.

PARSE VAR x1 w1 w2
           w1   ->   '<.A.B>'
           w2   ->   '<. ><.E><.F><>'

PARSE VAR x1 w1 w2 .
           w1   ->   '<.A.B>'
           w2   ->   '<.E><.F>'

The word delimiting allows for unnecessary SO and SI to be dropped.

x2 = 'abc<>def <.A.B><><.C.D>'

PARSE VAR x2 w1 '' w2
           w1   ->   'abc<>def <.A.B><><.C.D>'
           w2   ->   ''

PARSE VAR x2 w1 '<>' w2
           w1   ->   'abc<>def <.A.B><><.C.D>'
           w2   ->   ''

PARSE VAR x2 w1 '<><>' w2
           w1   ->   'abc<>def <.A.B><><.C.D>'
           w2   ->   ''