With xslt you can find the PIs <xsl:template match="processing-instruction()">, but parsing the content will be up to you. You could grab/be inspired by this template from the docbook xsls:
http://docbook.sourceforge.net/release/xsl/current/doc/lib/pi-attribute.html which parses the content of a PI to get the value of things that look like attributes, e.g. in author="dcramer" <?xm-deletion_mark author="dcramer" time="20090211T145246-0600" data="including"?>.
For manipulating the insertion marks, xslt 2.0's grouping features would come in handy.
For learning xslt, Jenni Tennison's
Beginning XSLT 2.0: From Novice to Professional would be a good place to start.
David