What exactly is so great about those benches?
1) They're far from objective. They compare nForce to AMD 760 in StreamD, but leave it out when comparing the Sysmark and Business Content/Content Creation Winstone benchmarks. The excuse that they're comparing integrated chipsets doesn't cut it either because the i850+P4 used in the Winstone benches isn't an integrated chipset solution. If you look at Anand's review of the KT266A, you'll see why nVidia at this point isn't showing real world benches comparing nForce to ANY other Athlon chipset.
2) They don't mention what version of the benchmarks they're using. As for "that Italian magazine," that's the Italian version of PC Magazine, not some fly-by-night publisher. Assuming the benches in the article used the 2001 versions of Winstone, PC Professionale's benchmarks make nForce look better. I'd also take the PC Professionale's benchmarks of a production board as being more valid than nVidia benchmarks of an nVidia reference board.
3) The 3DMark benches are useless. Anyone that knows anything about integrated video knows both the i815 and KM133 stink at gaming, particularly at higher resolutions. A much more interesting bench would be a SiS 735/AMD 760 board with a GeForce2 MX 400 to see how efficient nForce's architecture is.
4) All that talk about the efficiency of the integrated LAN may be useless. Large vendors will stick with external NIC chip solutions because it makes support easier. The MAC has been in VIA southbridges for a while, yet almost no one uses VIA ethernet even though it would be cheap to implement it. Check out the "useless" Italian magazine. Their specs show that Asus is using a Realtek 10/100 controller, which is no real surprise. Same thing for Gigabyte, who is using a Realtek on their SiS 735 board even though the 735 has an integrated ethernet MAC.
[Edit] Plus what Buz said [/edit]