Digilent NetFPGA board, basically plug a bunch of gig-e connectors into a FPGA, upload some ethernet software and away you go. Upload a softcore and run spanning tree on the softcore.
You won't like the price of the board I mentioned. On the other hand, nobody said a dev board with a bazillion extra features you don't want, made in extremely low quantity, is the cheapest possible way to stick a bunch of ethernet PHY to a FPGA (and you could optimize the size of that FPGA if you wanted...)
The point of that provided example multi-ethernet FPGA development board isn't that its the best you could do financially or technologically, but that if you tried to do your own thing and screwed up, its probably difficult to do worse.
As a practical matter having fooled with much smaller and simpler things the price should end up competitive in the end.
The biggest problem is synth takes a long time, god only knows what the CIA and KGB ops embedded in the source, and the IP licenses probably make distribution of the source rather difficult. Its mostly a business/government problem, not a technological problem.
Switches have sophisticated hardware on both low and high ends, but many widely used entry level L3 switches are essentially nothing more than large amount of ethernet NICs connected to single CPU (sometimes with hardware acceleration which usually means specialized DMA engine). This is even more true for routers (some Cisco platforms even have various acceleration hardware that their software simply does not use).
