There is overhead associated with DESTROY that doesn't exist (in quite the same way) for hash-based objects where garbage collection is left to Perl rather than done by hand. If lots of objects are being generated and destroyed, this may be an issue for some individuals.

Although when I've benchmarked inside-out vs hash based versions of objects in real world scenarios the overhead of the extra DESTROY just disappears in the noise.

I suspect it's going to be a rare environment where one extra method call perl object makes a significant difference.