Hi John
When data is preprocessed with meica.py, we find that the defaults of kdaw=10 and rdaw=1 works stably across a wide range of datasets (7T, multiband, 11.7T rat, kids, adults). What values are you using? kdaw>20 will likely lead to some instability. Datasets processed with kdaw=10-15 should be comparable - you can show this based on similarity of total variance explained by the ICA.
I still strongly recommend preprocessing with meica.py because it handles issues like dataset scaling and masking for multi-echo data correctly. Accurate component selection depends on proper preprocessing as well, so be careful if you're using a variant preprocessing pipeline. I'd be curious to hear a bit more about your implementation if you want to discuss it.
-Prantik