The speed of IP::Country::Fast is dependent on your processor and version of Perl. To look up any IP address, it takes a maximum of 32 x substr and 32 x unpack, with a couple of concatenations and arithmetic ops. These are extremely fast operations, and the cost of that code is pretty much zero.

However, on the first call to IP::Country::Fast, the module reads a 320KB file into memory. On the first use, this will involve disk access, but on subsequent uses any modern operating system will have cached the file in memory.

As a web service, any overhead from the module will be dwarfed by other parts of the system.

IP::Country is deployed on some very large sites, mainly used for geographic redirection, but I know it's also used in SpamAssassin and log analysis software. The underlying database is also used in JavaInetAddressLocator, which is bundled with a couple of Java content-management systems.

I receive a couple of emails a year from people who ask why a particular address isn't included, and when I receive such an email, I figure it's time to release an update to the database. The country code data is very, very reliable, but I have no real data to back up this assertion as I haven't used IP::Country myself for over five years. It's now just something I update and other people use.