I think it's worth noting that for this to work reliably, you'd have to make sure that the web server side end of FastCGI (such as mod_fcgi, mod_fastcgi) is configured to run only one instance of the respective process.  Otherwise, different requests could happen to be routed to different processes, which would produce unexpected results. Not to mention the confusion that would result from different people using the same CGI program and state variables.  The same problem would also occur with several apache/mod_perl instances running...

In other words, the easiest (and proper) way to achieve persistence across requests is to use sessions (CGI::Session), which makes your data being stored/retrieved independently of the process handling the current request.