I haven't read merlyn's column but here is a quick simple option (simple is relative)...

First, you could have the first page spit out a meta refresh tag that will bounce to the second page after a predetermined amount of time, if the second page isn't yet ready you could print another meta refresh. Alternately you could have a pipe open between the two processes, the forked one could could signal the parent, via the pipe, when it is ready, the parent would then send a javascript location bounce or a meta refresh to the browser (most browsers now will even let this occur at the bottom of a page, not just the headers) that will take you to the second page.

Second, the forked process prepares the html and dumps it to a static file under an assumed session id. The meta refresh or javascript bounce will just need to pass the id along. When the browser hits the cgi again, the second page simply opens up and spits out the cached file.

I haven't looked but I would imagine merlyn's solution does something similar.

my @a=qw(random brilliant braindead); print $a[rand(@a)];