Unsolicited code review:

    system("multimarkdown -s ../#{part}/story.md | sed -E 's/_([^_]+)_/<em>\\1<\\/em>/g' | sed -E 's/<h1 .+<\\/h1>//g' | sed 's/<p>%<\\/p>/<p class=\"section-break\">%<\\/p>/g' > output-#{part}.html")

IMO:

    s_<p>%</p>_<p class="section-break">%</p>_g

is easier to read than:

    s/<p>%<\\/p>/<p class=\"section-break\">%<\\/p>/g

Any reason to keep the %-sign?

Those sed commands could probably be reduced to one invocation of sed(1) instead of three.

    system("cat #{htmls} | #{wkhtmltopdf_cmd} --footer-html footer.html - body.pdf")

No need to use cat.