For what it may be worth, this kind of practice makes me very nervous.   Not so much (in this case...) that someone might take nasty advantage of such things, but rather, that it might be The Devil Himself to debug and to maintain on an ongoing basis.

I would consider tackling these differences through the use of Perl classes... a base-class that handles the task for log-files in general, and descendent classes which provide the details of particular file variants.   Each time you need to extend the application to handle a new twist, you can do it (robustly and safely) by creating a new descendent module.

Now your configuration-file would basically just have to say which class needs to be required into existence.

My admonition is, “an arrangement such as that, could be made to be rugged, whereas one based on tricky evals might be a constant source of headache.”   Even if, in the first blush, the latter seemed easier to write.