By far, one is able to get better noise reduction if you apply it to the raw sensor data, rather than after the raw data has been de-mosaiced, adjusted for gamma and transformed into numbers in some color space, etc. This is one of the reasons people why love the NR built into the last few versions of ACR and LR. One can explain this in terms of statistical signal estimation and related theories, but the bottom line is that experience has proved that NR software unequivocally works better if done as early as possible in the signal chain, rather than later.
For this reason, if you want better NR than ACR or LR gives you, I would strongly suggest you look only at RAW converters with built-in NR. Although I have not yet had a chance to demo it, several photographers that I respect are saying that the "Prime" setting in DxO Optics Pro 9 is currently the best-of-breed ... by a big margin. Of course, there is a downside: a) It's something else to spend your money on (LOL); and, (b) you need a powerful computer to run it at other than a snail's pace.
With respect to post-raw-converter noise reduction, there is a much wider choice. Each one that I have demo'ed has strong and weak points. For example some will be better at popcorn like small area noise (aka, high spatial frequencies), whereas others will be better at the sort of correlated grain that comes out of film scans. Some will do better in one tonal range, not so good in others, etc., etc.
Because of this, IMHO, there is no single post-ACR NR SW that "does it all". I have an absurdly large number of different types of NR SW on my system, but, if the NR in ACR doesn't work well enough, the next one I'll try usually is either Neat Image or XiMagic.
HTH,
Tom
