I don't think anyone's suggesting it's a magic bullet, but it can be very powerful in the right circumstances.
Let's take a real-life scenario just for example... Site rolling out new web-based product over https with 128kbps footprint per user. Site has 100Mbps link already at 40% utilisation, and expects 700 concurrent users of new app. The figures result in a saturated line. Could upgrade site link, but backhaul only 100Mbps from non-21CN POP so waste of time.
Riverbed benchmarks show 60% reduction in effective throughput due to caching of dupe data at site and takes site back within available bandwidth.
Can be useful in the right circumstances, but an expensive experiment if you don't do your homework and validate it as a solution for your requirements.
So as you say you've done a comprehensive eval, how did it compare against the built-in de-dupe tools in Windows Server 2008/2008R2 ??