I assume you're talking about hosting this VPN with Microsoft's Routing and Remote Access (RRAS) server.
I'll preface by saying that I've played with such a configuration but don't have any such configurations running in production.
You can assign a pool of IP addresses outside the subnet of the RRAS server to be given out to VPN clients. When I played with this in a lab scenario I found that the RRAS server (on W2K8 RTM, at least) would start doing crazy things like answering requests to its LAN interface with packets sourced by the RRAS interface (which isn't on the same subnet as the LAN interface). If you can get a route into the router acting as a default gateway for the VPN subnet then things should be fine.
Admittedly, it's a bit of a strange configuration. Generally, RRAS assigns IP addresses from the same subnet that the RRAS server is sitting in (typically via DHCP) and does proxy ARP "magic" to make those clients appear to be on the same layer 2 as the RRAS server's interface.
When you're mocking this up, throw a copy of "Network Monitor" or Wireshark onto the RRAS server and actually watch what's happening with the source addresses on packets. I think that'll give you a clear indication of where you need to put routes to get what you want to work.