Re: a better web platform, one way to get ahead would be to solve the C10K problem - http://www.kegel.com/c10k.html. The whole CGI/FCGI thing is a dead end. In-process solutions like mod_php and mod_perl take you a little further but not far, unless you're 100% thread safe, and even then, there's a limit.

Solution: a Pure Perl web server based on async IO, with a decent API, and relevant modules (e.g. session handling) bundled.

And decent API does not mean CGI-like - see http://www.mnot.net/blog/2003/12/08/http_api. It also doesn't mean Catalyst, which inverts the problem to "many verbs via a single controller". Check out web.py's API (http://webpy.org/). And see STREST - http://duncan-cragg.org/blog/post/strest-service-trampled-rest-will-break-web-20/

Practically perlbal seems to be the best candidate (http://www.danga.com/perlbal/) - repurpose as an application server and you've got an edge over the competition, for everyone that wants serious scaling. Sure it's not a solution for the low end user but right now there's effectively nothing that a high end user can take seriously, other than Java. Perlbal has already got async IO going on plus it's served enough requests to be regarded as fairly robust. Someone has already thought about it: http://lists.danga.com/pipermail/perlbal/2006-May/000205.html