Multibib reverse label or sort order

The biblatex package offers features comparable to those of multibib. References from a single bib file can be subdivided using various options of \printbibliography. Examples include type, keyword and category. Refer to the biblatex manual for details. Prefixes to the labelnumber are specified with the prefixnumbers option.

Assuming each sub-bibliography has a unique prefix, descending label numbers can be obtained by altering the labelnumber field format so that it prints the value given by the total prefix-specific entry count, minus the actual labelnumber, plus one.

\documentclass{article}
\usepackage[style=numeric-comp,sorting=ydnt,defernumbers]{biblatex}

\AtDataInput{%
  \csnumgdef{entrycount:\strfield{prefixnumber}}{%
    \csuse{entrycount:\strfield{prefixnumber}}+1}}

\DeclareFieldFormat{labelnumber}{\mkbibdesc{#1}}    
\newrobustcmd*{\mkbibdesc}[1]{%
  \number\numexpr\csuse{entrycount:\strfield{prefixnumber}}+1-#1\relax}

\addbibresource{biblatex-examples.bib}

\begin{document}
Filler text \cite{aksin,bertram,angenendt}.
Filler text \cite{chiu,padhye,moraux}.
\printbibheading
\printbibliography[prefixnumbers={A},type=article,title={Articles},heading=subbibliography]
\printbibliography[prefixnumbers={R},type=report,title={Reports},heading=subbibliography]
\printbibliography[prefixnumbers={P},type=inproceedings,title={Presentations},
                   heading=subbibliography]
\end{document}

This solution works with either backend, though with biber the "rerun LaTeX" messages in the log may not stabilize. In any case heed the warning; the first time the message disappears after recompiling should give the correct document.