The user gets to decide, by opting to use the module, or to not use the module. Minimum version requirements will weigh into the equation for some portion of users. The more modern the requirement, the more users will find themselves among the portion where minimum Perl version is a factor.

Authors get to decide where to draw the line in the tug-of-war between hubris wanting to include as many potential users in their club as possible, and laziness wanting to avoid the work involved in providing a solid product that also works for older versions of Perl. (I'm trying to use the terms "hubris" and "laziness" in the Perlish sense, where they are both desirable attributes, and not character flaws.)

Sometimes the issue is decided by bugs that were fixed in newer versions of Perl. (Sure, an author could probably find a work-around that doesn't cause the regexp engine to segfault under 5.8.4, but the work required may tip the scales for him.) Sometimes it's decided by improvements to Perl that make hard things possible, where they were once much harder. And sometimes hubris degenerates into the form of ego, manifested by simply not caring that the only reason 5.10 is necessary is because it's more idiomatic to write //= than a more verbose construct. The author decides where he finds personal balance. The user decides whether to use the module, find a different one, or reinvent.

Either way, both the author and the user have a vote.