Maybe not bad, but not good either. I'm reminded of some of the underexposed test shots I took with my 5D a couple of years ago. Here's one that I consider to be an ISO 51,200 (ISO 3200 underexposed 4 stops). It's seems about roughly as noisy as the D3s RPP conversion of the ISO 102,400 shot, which would give the D3s a full stop better high-ISO performance than the 5D. The D3s shot appears a bit sharper, but that might be due to the that fact my shot was hand-held at just 1/25th sec.
The screen grab of the ACR window shows a +3 EV adjustment which might give the impression the shot is only 3 stops underexposed. However, my reasoning is a full ETTR exposure normally requires a -1 EV adjustment in ACR to get the histogram looking as it does in this shot. There are no values anywhere near 255 and the second image showing the levels histogram after conversion, indicates there's plenty of highlight headroom, so I'm convinced this shot is at least 4 stops underexposed at ISO 3200, but I'll listen to reason if anyone wants to persuade me otherwise .
[attachment=17530:Canon_5D...R_window.jpg] [attachment=17531:Canon_5D...onverted.jpg]
I know it's difficult and inconclusive to compare image quality using completely differenct scenes, but I prefer the 5D image in this case, probably because I'm closer to the subject. The D3s shot levels histogram shows a lot of black clipping. The slight clipping of the highlights is due to specral highlights which are not present in my shot.
[attachment=17532:RPP_conv..._102_400.jpg]