1. How do I access the subnodes for expressions like "S: A*;" ?

In this case, "A*" is $n0 of type D_ParseNode.  In dparse.h the functions d_get_number_of_children(D_ParseNode *) and d_get_child(D_ParseNode *, int) allow you to get the number of children (A's) by calling d_get_number_of_children(&$n0) and individual children by calling d_get_child(&$n0, x).

See the example: tests/g27.test.g

2. Why doesn't '$n0.end_skip' include the trailing whitespace when $n0 is a string/regex?

Whitespace is not skipped as part of the scan of a string/regex but is done later (when the new parse state is created).  This makes it possible to change the whitespace parser as a result of recognition of a particular string/regex.  This is used in the python grammar to handle implicit line joining.

See the regex-productions LP RP LB RB LC RC in test/python.test.g.

3. What is the difference between DParser and ANTLR?

The basic syntax of DParser and ANTLR grammars is very similar.  In fact, the DParser example ANSI-C grammar was ported from ANTLR in less than an hour. 

Beyond that there are a number of differences.  First of all ANTRL has been developed for over a decade while DParser is relatively young.  Internally, DParser is a table-driven parser while ANTLR generates directly executable parsers.  DParser is a GLR parser based on the Tomita algorithm while ANTLR is modified LL(k).  DParser is scannerless while ANTLR uses token streams.  In terms of power, both DParser and ANTLR are very powerful.  In theory DParser can handle any context free grammar, though not necessarily in linear time.  In practice ANTLR is likely to be faster (mostly because it is more mature).   Beyond that you would really have ask Terance Parr who is more of a hard core parsing theory guru than I am.