They don't show the raw CPU performance against anything non-Apple, but I'd guess they'd perform roughly the same as any other dual-core A9 at the same clockspeed (with the caveat that Tegra2's don't have NEON), though most Android phones have higher clockspeeds. Since 1GHz Tegra2s from tablets beat it even on Sunspider it seems they still have a raw CPU disadvantage due to clock speeds.
Everyone and their mother knew the SGX543MP2 was gonna smoke everything else on the market right now. Maybe the new Nexus will leapfrog that but as of this week "beats all smartphones" in benchmarks would seem accurate.
As I said, there's no straight CPU benchmark, or anything that is limited by RAM so ""beats all smartphones" in benchmarks" is a fair bit less accurate and complete than what I said.
There's also the fact that Hacker News generally frowns upon editorialized headlines. What was so wrong with the original title of "iPhone 4S Preliminary Benchmarks: ~800MHz A5, Slightly Slower GPU than iPad 2, Still Very Fast" that it needed changed?
(edit: getting very pernickity now but since you misunderstood my comment about comparing the sunspider benchmarks: Android's javascript loses on the sunspider benchmark to iOS's if run on the same device. So if a 1GHz Android device (without NEON) matches a 800MHz iOS device then it's quite likely that a 1.2GHz phone chip will beat it by a fair margin in Sunspider with the Honeycomb or ICS V8, by more again on other Javascript benchmarks which it generally does better on, and more yet again on anything that just benchmarks raw CPU power. I keep reading about Apple's magical A4 and A5, so I'm being pedantic on this point, but the gain is that they chose to use a better GPU from PowerVR and updated their javascript VM, not pixie dust or other Apple magic used in their hardware).
