> If the only places that forwarding speed matters were in high end 
> routers with FPGAs, ASICs, and similar, there would be more merit 
> to that argument.  But fast routing (if not *that* fast) is desired 
> by other systems too - like the average box running NetBSD, and what 
> Tony is able to get a FPGA to do, isn't all that is relevant. 

What fraction of the CPU cycles on a NetBSD router are spent in the 
routing code?  How much does that differ between IPv4 and IPv6?