for me the biggest drawback of rx100 is control wheel, it needs two Big impulses to change value, say exp. comp., first just displays the info on LCD. bummer!!

First of all they don’t measure noise using image with real content but completely flat, detailless, batches of various luminosity.
While excellent for easy dynamic range measuring it has serious drawback:
It doesn’t show in any way destructive effects higher noise has to content of image!
(Hands up everyone who shoots only flat white/gray walls!)

And clean and simple theoretical averaging of “random” noise away by downscaling image doesn’t work so easily in real world because noise isn’t random from pixel to pixel.
During demosaicing every pixel has two of its colour values interpolated using surrounding that colour sensing pixels of sensor.
So higher noise/lower signal quality in single pixel of sensor affects to surrounding pixels of image lowering also quality of those and increasing noise grain size bigger than pixel.
More complex demosaicing algorithms look even farther than immediately surrounding pixels when interpolating missing colours so noise probably has tiny effect up to two pixel radius.
As consequence downscaling factor needed for certain level of noise decrease is higher than in theory.

If you want to find out how well averaging noise by downscaling works in real pictures following test wastes less time than reading DxO’s theoretical jargon about how you can make party cake out of sh*t if you just have enough of it.
Go to Imaging Resource’s site and open reviews of both Canon Powershot G10 which had 15MP marketroid sensor and its successor G11 with 10MP sensor.
Then download ISO1600 still life shots of both. While G11′s image isn’t exactly clean it’s lot better than G10′s heavily NR blurred but still noisy picture.
Now simply downscale both images heavily (download zipped XnConvert for trying different algorithms) to 1600×1200 to see G10′s image being still clearly uglier.

Dpreview has excellent, very easy to use RAW comparison tool and in overall E-M5 has nothing to shame to APS-C cameras. (winning Canons)
Again Imaging Resource checks how big prints can be made from various ISO photos, and got similar results.

And Olympus didn’t have different 12MP sensors but just same old mediocre already in 2008 design. Improving read out speed for faster contrast AF did in no way improve photosite/pixel noise performance. Same for slapping lighter AA filter in front of sensor.

The EM-5 hasn’t been tested by DxO yet but it will likely be a bit better than the best m43 results to date. I suspect it will likely place around first generation NEX sensors but well below the NEX-C3, 5n or 7, perhaps around 850-900 on DxOmark’s low-light ISO score:

Best APS-C tested by DxOmark: Nikon D7000 @ 1167 ISO
Best m43: GH1 @ 767
Best compact: RX100 @ 390 ISO

As there’s much greater parity between the best representatives of formats in DxO’s dynamic range and color depth ratings, the basic assessment camera buyers have to make in sensor format selection is really how much low-light ISO and DoF subject isolation they are willing to pay for and lug around. The RX100 really destroys the case for the Nikon CX format, but doesn’t touch upon those two reasons for going still larger.