Boris,
I am a bit confused. But since Markus was able to generate clean static files at his place and you get the same noise at your place, it is highly likely that you have some sort of local interference. Other people have had to move location or change the setup (new type box, insulation etc) alltogether to get acceptable static files...
Also, looking at the Z-mag in your dynamic, it is a bit "fuzzy" when static, this could again point to interferences in the environment or (more likely) interference from some peripheral or the power distribution on the frame. But its not too bad - something to look out for though..you should be able to generate a result with it.
Regarding step 4a and following steps:
If the MAE is jumping you can try and run that step with a lower variance. From the wiki:
You may also need to change the variance parameter (to the left of the calculations window, above the ‘Start sim‘ and ‘abort‘ buttons. Default variance for Sim3 is 1e-6 (0.000001) so try 1e-7 or lower. Check the ‘use variance‘ box and enter a number in the field below, then hit ‘Start sim.’
Sometimes it can be good to run a few dozen/hundred loops at a low variance to converge the data and save the results into your params file. Then run the step again at a higher (or default) variance. The more you lower variance the longer it will take to arrive at a good solution.
In general, if you have to lower the MAE throughout the Sim3 calculations, it points to problems with your data.
Noise from movements in the beginning and end of the files should be cut away. Looking at the dynamic files, you have a big spike in the end of the gyro and acc plots. The dynamic file should have a static period in the beginning and end of the file. (craft on level ground, not disturbed for 10-15 seconds)
Otherwise, you are definately heading in the right direction with the dance - its looking smooth and slow enough to me.
Look at the MAEs produced for gyo and acc. MAE for file 1 is much higher than for file 0 and 2 - and the same goes for accel run. Id take a closer look at that file and see if it has noise in the curves.
You can produce a result with just one dynamic and one static, if they are good quality. So you should be able to get away with just using static file 0 and 2.