Forefront TMG 2010 … UPnP Port mapping

There's no comparison because they're really two totally seperate products that are aimed at two different markets. Kind of how you'll probably never see a comparison of a Ferrari 599 against a Bugatti Veryon. Both crazy fast expensive cars, but aimed at two different markets.

I've used both. In fact our internal office uses TMG, and our remote site uses PFSense, and it really comes down to what you're after in a firewall appliance and your ability to maintain.

I find PFSense a breeze to maintain. Setting up failover links, IPSec tunnels, VPNs, etc very very simple. All of this is much more complicated in TMG, but that's because TMG is very tightly integrated into your Active Directory environment.

TMG can also do host-based HTTP routing, whereas PFSense can't, so you can use one IP address across multiple internal web servers without needing a specific reverse proxy.

One of the best things about TMG is that you can effectively turn off just one person's internet access by disabling their AD account in the firewall. No need to set up any Squid authentication with RADIUS against your AD and then setting up ACLs.

I'd say the TMG is more difficult to learn then PFSense if you're starting from a blank plate.

NLB relies on nodes being on the same broadcast plane as each other, so doesn't usually work well in geographically distributed cluster scenarios (that, and it often costs you a lot in bandwidth broadcasting all incoming traffic across a WAN, only for it to be dropped over there).

Other clustering - perhaps, depending on what you have in mind. TMG doesn't lend itself to clustering except with its integrated NLB, so it'll have to be above (or below) the TMG level.

Having another VM ready to go in the DR site, and using DNS to direct traffic to the curent live node for internet clients, is probably the best solution. You could install this as another TMG Array member if it's used exclusively for publishing, and not enable NLB; that keeps the configuration consistent between the nodes, and allows for externally-driven failover.