About Reinventing CGI
CGI is a library which has been thoroughly tried and tested and if it needs improvement, you could build an extension, or contribute/contact the maintainers. You have to remember that modules are only as good as their track record (reliability) and their upkeep. Many people created decent modules, but didn't continue maintaining it, so they sort of fell to obscurity.
CGI is a boat all of its own, which if you think there's a lot of overhead, you could use CGI::Simple or CGI::Minimal. CGI.pm does more than parse querystrings, it also has cookie management (sessions), HTML generation, and other useful functions.
Others have had some criticisms of the overhead with CGI.pm, but that's why they developed FastCGI, which is modifying the server to use a persistent state of the script, thus loading the overhead once, rather than on every page load.
It is possible for you to create another (even better) version, but why bother? Many people may probably tell you, you shouldn't reinvent the wheel, with good reason. CGI has been around for almost 2 decades, with so many users testing it, finding holes, and having patched the holes; however, I'm never a big fan of saying "you shouldn't do something." If you think something could be better, make it better. There are many OSes that exist today just because of that reason, why settle for something that does 95% of what you need, if you need that other 5% too? But I also say, weigh your costs vs. benefits and determine if you want to devote your time to this, or if maybe there's another problem out there that has yet to be solved, that could use a little more manpower. To have something successful, you're going to need to test it thoroughly, and will most likely need to create something that other people would want and (at this point) there isn't much of a reason for CGI-users to be motivated to switch.
About Modern Perl
I think "modern Perl" is an oxymoron. I would jokingly call modern Perl; Ruby or Python.
That isn't to say that Perl isn't useful, because it is, but it's been around a long time. While it has had its significant share of changes from version to version, the most popular, Perl5, hasn't changed all that much; mind you, my definition of change is not adding to the language (new operators and functionality), but deprecating/replacing old features or changing the behavior of existing ones (like for
/foreach
loops).
Note: Perl6 could be considered modern perl (and does have many significant changes), but it's not widely adopted and was supposed to be released many many years ago (it's the Duke Nukem 4 Ever of programming languages).
About XS
I haven't done much module programming, but if memory serves correct, XS is the interface between Perl and C, which I think allows you to compile your perl modules for faster execution. Consider the PostgreSQL DBI module. There's a DBD::PgPP, which is a pure perl module to interface with Postgres, but there's also DBD::Pg, which I think compiles some of the code using C and takes advantage of some other OS utilities. Compiled modules have the benefit of faster load and execution (there may be some better resource management in there too).