MY PROBLEM SEEMS TO BE SOLVED.
For your information, I report here the results of the analysis and the solution we arrived at.
As per previous posts, I confirm that the problem was due to a single user with an anomalous number (hundreds of thousands...) of files in the "AttachedFiles" folder, due to endless recurring appointments that contained an iCAL (.ICS) attachment.
These "anomalous" appointments were generated by an invitation to meetings created with Google Calendar and Google Meet by an external user (the external user is the Owner of my client company, which uses Google because he also owns 20 other companies that all use different systems and therefore he is fine with it. Oh well...).
Of course you understand that I CANNOT delete the appointments...
The final solution was to do this:
1 - export the offending appointments in .ICS files via the Webmail menu. This way everything is saved EXCEPT the problematic attachment.
2 - delete the appointments on the webmail (the whole series). This way the "AttachedFiles" file repository is cleaned
3 - re-import the appointments from the .ICS files previously saved in point 1. This correctly recreated all the recurring appointments with all the details WITHOUT the iCAL attachments that were weighing everything down.
NOTE: to find any users who have these problems you can search the "error" log for reports similar to this one:
<<<<<<<<<
[FileFunctions] File is over 15mb, this might cause reduced performance. Size: 184.8mb Filepath: S:/SmarterMail/Domains/DomainXXXX.com/Users/UserXXXX/AttachedFiles/attachedfiles.json
>>>>>>>>>
In the ticket I opened I was assisted by Ray Burd, who confirmed that in the past (v. 9084 approximately) there was a bug that could create these problems with attachments from invitations received from Google, but he says that in the latest versions it should have been fixed.
The problem was "venial" if you received an invitation for a single event, but if the event is infinitely recurring, then it causes the RAM leaking that I saw. Let's hope it is truly solved...
Of course I asked him to do further checks on the correction of this BUG!
P.S.: Thanks to Ray and Tony for the help!