We run a shared hosting webserver with the usual LAMP stack. It is up and running since many years ago (uhm, Apache-1.3 and PHP-3 days?) and went through many iterations. We strive to have good sysadmin policies, like keeping all the stack up-to-date, checking for weak passwords, minimizing the attack surface, using suhosin, keeping an eye on systems logs, and so on. Of course every virtual host is confined to its directory (both for FTP access and as php open_basedir).
But at the end of the day it's always a webserver running untrusted PHP crap uploaded by customers (read: unknown and mostly-stupid users without any IT experience) with HTTP exposed to the world (web forms and whatnot)... these scenarios are not too uncommon:
- user gives its password to too many people and its site gets compromised
- user's PC gets compromised and the FTP / web app / whatever password stolen from there
- user installs crappy PHP stuff and it gets compromised
- user installs good PHP stuff (does it even exists?) but doesn't update in years and it gets compromised
- user writes its own PHP stuff (ARGH! they're coming outta the ---------- walls!)
- and so on. you got the idea.
With such a wild environment it is not possible to repeatedly run automated vulnerability tests on the web sites, and probably it wouldn't be much useful either. I'm also thinking stuff like mod_security would be out of question in such a shared, generic, out of control environment.