JDocs :
Off the top of my head here so I'm probably wrong.
I've read that the 1.8ghz Atom is about the same as a 900mhz P3 in rough terms. That indicates that the Atom's instructions per cycle is half of a P3. A P3 does 2 (again off the top of my head, based on the P4 IPC since mhz for mhz they where similar). That would imply the Atom does 1 instruction per cycle. Core 2 65nm does 4 IPC and Core 2 45nm does 5 IPC. Effectively a Atom dual core @ 10ghz would rival a 2ghz Core 2 45nm....
Again, off the top of my head so please no being nasty...
Correction A p3 like a p4 and does 3 instructions per clock, a p4 is 20% slower than a p3 per clock.
(I had a p3 Tualatin Celeron running 1.6 (and another one 1.8 for a short time) with a 33 to 36% overclock.
An Atom does either 1 or 2 per clock, I don't remember anyway this number does not directly give you the IPC (Performance per Clock) there are a lot more variables like pipeline depth, architectural differences, cache. cache speed, busspeed, etc etc
And this is just plain wrong: A Core 2 45nm does not have 5 IPC, a Core2 45nm does the same amount of instructions per clock as an 65 nm, the extra cache (and in rare cases the SSSE3 and SSE4_1) gives it a higher IPC but that is not a number like 4 and 5.
You are over simplifieing IPC and confusion it with I think pipeline width.
As long as there is no bandwith requirement like fast ram the P3 will outperform a slightly higher clocked p4 anyday.
(Of coarse with the later P4 with sse2 and sse3 and newer software the comparison gets trickeyer.)