I have tried it two years ago. It had many severe limitations, but was usefull as an addition PL/pgSQL for manipulating text. The main limitation is that you cannot use your database tables directly in your Perl subroutines. Second was that using libraries broke the security model. But that was 2 years ago - so it might have changed.

Update: I read the linked PL/Perl documentation and indeed the situation looks much better than two years ago with the experimental DBI::PgSPI solving the first issue and more convenient use of plperlu for the second. I can see much potential in this kind of database programming, and perhaps when more people start to use it it shell become less 'experimental'.