LOL or think smarter. Performance advantage allows you to do any amount of muscling you want and at 22nm all x86 (both Core and Atom) processors will have a performance advantage over ARM which they will never then lose going forward.
If we take Intel's claims of 60% performance improvement (@22nm) over Clovertrail at face value and we use Geekbench as our performance test suite then it's unlikely that Intel will have a CPU performance advantage over ARM at that node (Intel Atom 22nm V foundry 32/28nm).
It is scaremongering based on ignorance and convenient forgetting of the results of decades of cpu technology wars. Cyrix, Transmeta, AMD all tried this cheap low power cpu approach to combating Intel and they are all either dead or dying. It just doesn't work if you don't have the performance or features to back it up and ARM doesn't, like all the RISCs that have tried it will ultimately fall short against the CISC x86.
Cyrix, Transmeta, AMD and co all had to implement the x86 ISA in one form or another. ARM don't so they aren't spending their transistor budget on legacy instructions to maintain binary compatibility with Intel. Currently, ARM's CPU's are much smaller than the equivalent x86 core and thus have a die size advantage. Clearly this will change going forward, especially as GPU's are becoming the largest item on a SoC . I am not too sure what features you think the ARM ISA is lacking?
The CISC/RISC argument:) RISC actually won in terms of design, as all Intel's Core processors are really RISC with a CISC decoder sitting on top. Regardless, over the years the RISC/CISC ISA's have moved closer together so is a mute point. I suspect you are talking about how x86 killed the RISC server ISA's?
If you are using Geekbench 22nm Atom will have a performance advantage in Integer, core vs core, and Integer is the most important metric to consumer apps. Geekbench is fine if you don't have other more common applicable consumer app benchmarks but there are many out there. Geekbench is like a poor man's SPECcpu and should be used as such, a fallback comparison metric when you don't have more relevant benchmarks. The CISC/RISC distinction is still important as you can attempt instructions and operations in CISC you just could not do in RISC. The most glaring feature ARM is lacking is 64-bit at the moment and as you just don't get consumer o/s that use PAE that will limit memory to 4 GB and non-o/s user memory to 3 GB or less. x86 also beat PowerPC in Macs, Alpha/Sparc workstations as well as ARM the first time round, 486 vs Acorn Archimedes in addition to the general mauling it has dished out to Server RISCs.