Hi,
We have operated this way for a number of months. If you are using NAT, you just need to define an auto NAT on the Policy Server objects and ensure the NAT policy is implemented on the firewall in front of the Policy Server.
We have done full deployments using the installer agent in this model.
Things to consider:
1) The endpoints will always try to talk to the manager as a path of last resort, so you need to actively block this traffic if it is accessible via RAS etc.
2) The endpoints will try to connect to the native and NAT IP addresses so clients may traverse different paths depending on whether they are VPN connected or not. Again, we prefer to block the native traffic and make it predictably come via the Internet (also saves crypto overhead on RAS gateway).
3) If you perform remote upgrades, you need to take care not to swamp your internet connection as a large number of clients could do.
We have also asked Check Point questions around whether the Policy Server is designed to be Internet facing and also whether its possible to reverse proxy the traffic. So far haven't had a response. The "outer shell" of the server is Apache and a quick config review has shown it to be reasonably well hardened and configured. Of course you want to place it behind a properly configured gateway with up to date IPS protections etc.
Cheers