How exactly do you run Saxon? I think it might depend on the behaviour
of the underlying parser and I am not sure Saxon HE, as it itself does
not support schema-aware XSLT, uses any xsi:schemaLocation hint or
passes it to the XML parser, so basically the schema is ignored, while
the default setting for XML based parsing might use a parser reading and
taking the DTD into account. I would expect Saxon EE with
schema-validation turned on for parsing to give a different output for
the second sample.

Specifies what whitespace is to be stripped from source documents
(applies both to the principal source document and to any documents
loaded for example using the document() function). The default is
ignorable: all ignorable whitespace is stripped. Whitespace text nodes
are ignorable if they appear in elements defined in the DTD or schema as
having element-only content.

I guess in your case the DTD defines element only content for the author
element and Saxon and/or the parser used the DTD while the schema, even
if it defines the same, was not applied during parsing.