Hi, glad to hear your script was improved. To your questions:

(1) Re: S_IFDIR. Easy to check: as I see there are 2 entries/directories in my C:/Users tree with FILE_ATTRIBUTE_REPARSE_POINT bit set; for both S_IFDIR is also set. So, to strictly follow Win32::LongPath documentation as to what to treat as "directory" i.e. exclude such entries, -- use fragment you quoted from documentation. Stress tests are good, but my impression was your tree to monitor should be known beforehand, whether it can or cannot contain reparse points. And, I meant the same thing (about "controlled environment"), when I said built-in readdir might be enough vs. readdirL, if a tree is known to be grown by native speakers/users of single (system) Windows code page. For versatility, sure, Win32::LongPath should be preferred.

(2) Re: <C:/Users/bo/Application Data/À> - I don't understand (was this a question/problem?)

About freezes: that's not good. I can repeatedly sweep C:/Users and C:/Windows without issues, both with my and your scripts. At first, I suspected iterative readdirL (furthermore interspersed with lstatL calls) might be the reason, but, no, your script runs OK here. Still, to debug, maybe try to switch from iterative use to list context call. BTW, I don't observe any noticeable speed difference.

Lastly, about different total results of (various) Perl techniques and/or what Explorer reports: if you really want to pursue to the core, there's dichotomy with extensive logging, but of course you know the method.