Before noticing your code is way too long, I spotted:

$value =~
      s/%([\p{IsAlphabetic}\d][\p{IsAlphabetic}\d])/pack("C", hex($1))
+/egxsm;
[download]

Now, sidestepping the question why you're rolling your own parsing, why match anything alphabetic (more than 100,000 characters will match) or a digit (more than 400 matches). And considering that [\p{IsAlphabetic}\d] is equivalent to \p{Alnum}, you could have saved a lot of typing. Of course, it's still wrong. Much better would be the pattern: %(\p{AHex}{2}), as \p{AHex} matches exactly the 22 characters you want to match. Not a single character more, not a single character less. Much better than the 100,520 characters you're matching.

And what's up with the modifiers? You have 5, of which you're only using 2. You've no characters in your pattern that will change meaning with the /s, /m and /x modifiers.