Once I made it through the setup, however, I found throughput to be surprisingly high. Figure 11 shows a composite Chariot plot of PPTP throughput test results in both Local-to-Remote and Remote-to-Local directions.
The PPTP client was a 1GHz Celeron-based notebook running WinXP home and using Microsoft's built-in PPTP adapter. Note that these results were obtained with no other traffic running through the router and nothing else besides Chariot and its endpoint running on the test pair machines. Although the 8-9Mbps performance is for a PPTP tunnel which is less compute-intensive than IPsec, I think it's still pretty impressive!
But since it's unlikely that you'll be running your network this way, I did a second test shown in Figure 12.
This test used two pairs of computers, with each pair having a machine on the LAN and WAN side of the router. I set the Chariot script to send data via a normal LAN to WAN transfer between one pair, and via a PPTP tunnel in the Remote to Local direction between the other. The results show the 8200 was able to handle both normal NAT and PPTP chores with plenty of throughput for each task.
On a negative note, however, I found that disconnecting the PPTP session from my WinXP Home client locked up the 8200 each time and required a power cycle to get the router operating again.
Since I couldn't get an IPsec tunnel running directly from an XP client, I wasn't able to really test IPsec throughput. USR told me they have tested over 150 simultaneous IPSec connections with "typical throughput scenarios" without seeing any serious degradation in performance. They also "maxxed out" the number of PPTP connections without any performance issues.
USR says the only hard limit on number of connections is 254 - the number of IP addresses in a Class C subnet. So unless you run a lot of large file transfers through multiple IPsec tunnels, it looks like you won't have any trouble with the 8200's VPN performance.