Very interesting question. My main applications in OSS are:

Library Lookup Helper (in-page status for checking library holdings)
http://webvoy.uwindsor.ca:8080/cocoon/mount/chrysalis/welcome

WIBS (Windsor Internet Booking System - originally based on MRBS but
recently rebuilt and the Linux client side is being tackled by a commercial
Linux solution provider)
http://wibs.sourceforge.net

PYTHEAS (ILS - not much activity in the last year, way too busy but have
given away code all over the place, I think koha was using this for some
ideas on marc mapping at one point)
http://www.uwindsor.ca/library/leddy/people/art/pytheas

Project FLOW (SVG and Google additions for Web-based OPACs)
http://webvoy.uwindsor.ca:8087/projects/flow

I graduated with a computer science degree in 1985 (I was using punch cards
with fortran IV in 1981, a fact which never fails to make me feel ancient),
and have written and shared code on all sorts of systems, including a
DOS-based TSR for logging out CD-ROMs and a SPIRES-based MARC editor (in
the late 80s/early 90s), a custom windows shell, and a java-based web
browser. The one that I have shared a fair bit of code over that surprised
me is a co-browser I wrote in connection with the Access 2002 hackfest
<http://137.207.184.23:8088/access2002/>. I have told anyone who uses it
that I would do it differently if I tackled it again (we didn't have time
for this in the hackfest and I wrote it out of sheer guilt soon
afterwards). Which leads to the other part of John's question, I sense
great interest in OSS applications for virtual reference, and I am fairly
convinced a killer application could be built in an environment like
Cocoon, since the proxy pieces for a co-browser are already there.

Another application in the "ones I will not easily get to" that would make
a lot of sense for an OSS implementation is a CIPA filter (see Cindy
Murdock's great story at WebJunction:
<http://www.webjunction.org/do/DisplayContent?id=3973>). I wondered about
using the common browser support of DDE (used by WebSpy and others) to step
into a browser session and then out again. One of the difficulties, I
think, with filtering applications is that they don't have a mechanism for
querying a user in a session and then providing unfettered access if they
get some sort of go-ahead from the patron (we don't deal with patrons, and
I live in Canada, so I may be really off-base on how CIPA is supposed to be
deployed).

Finally, a library-specific toolbar has come up several times, from such
notable commentators as Jenny Levine (The Shifted Librarian). I would
really like to see something like this done for lii.org, with a plug-in for
local libraries. The trick for me is to avoid platform-specific coding, and
toolbars seem to put you into the depths of the desktop.

So much to code, so little time...

art