Yes, more real-world tests are needed, for sure. I should dig up some of my high-ISO S1R wildlife shots.I remember all the debates about Topaz AI adding made up data, and real photographers would never use it. Despite that I've used Topaz Denoise AI and Topaz Photo AI quite a bit, almost always successfully (and now and then a case where it doesn't do a good job, it can throw away data it thinks is noise). I've not used DxO, or the DxO Lightroom plugin, so I can't compare it, but I will be doing some comparisons with the new Adobe Lightroom AI capability. A problem with Topaz is that it still doesn't accept S5II raw and you need to use the TIFF conversion path in Lightroom. This might give a slight advantage to the Lightroom version; but Topaz and DxO have a huge lead in denoise AI.
I remembered a shot with a deer standing on a road with fine gravel. Topaz AI would see this fine gravel as noise and clean it up so it looked like a dirt road. I went back to look at that shot again with Lightroom A1. It did a good job and left the gravel unchanged. But I looked at it anew with Topaz AI (which has gone through several updates since I looked at that shot last summer). And Topaz now looks at the gravel correctly - it leaves the gravel unchanged, so Topaz has fixed this problem. The results looked identical between the programs. A few other photos with noise also showed identical results with Topaz AI and Lightroom AI, both very good at reducing noise and maintaining detail.(and now and then a case where it doesn't do a good job, it can throw away data it thinks is noise)
Yes, I've been using Topaz noise reduction with m4/3 for quite a while. For some shots it makes a big difference. It doesn't turn a m4/3 camera into an S5, but it can definitely extend its range. When I have time I'll go back and look at some m4/3 shots I've recovered with Topaz and see how they do with Lightroom.might even make me rethink u4/3rds