After benchmarking all CPAN HTTP solutions and being a very long user of WWW::Mechanize; I ended up choosing Furl. Among 20+ contenders, it was among the top 5 in speed, and with a very low memory footprint. It was also the one who croaked the less over thousands of random urls.
Because it is lightweight compared to Mechanize, much more quicker, and it handle a lot of things automatically that Curl and others low levels HTTPer do not (Multiple Redirects detection hell anyone?).
Support with HTTP::CookieJar, nice simple interface, and just works. Documentation is a bit terse, but if you are familiar with LWP or Mechanize you should be fine. Used in production for external webscraping and as a Rest API call.