We're using perlbal for 2/3 webservers and will be rolling it out for 2 more shortly, it's handling between 200 and 300 concurrent requests despite sharing a server with a busy mod_perl server and proxying to other busy servers.
The reproxying is very very useful, the ability to daisychain selectors is also a killer feature.
So far my only issues have been the level of documentation, and occasional reliability issues under nasty conditions (i.e. when other applications/servers on the same machine were very busy).
However, if you need this, then you're likely to be experienced enough to deal with using mailing list archives, etc to answer your questions.