You may want to talk to IT from other companies to see what they do - I expect that allowing the webserver to access something inside the firewall is precisely how it happens. Through a VPN tunnel, or via very strict firewalls that only allow the one static IP address through, or by using a second network card that connects to the other side of the firewall, and using strict iptables to limit what can go through each network card. For example, on eth1, only transmissions to port xyz on machine a.b.c.d are allowed. This can severely limit the exposure should the DMZ box be rooted.