Your processing pipeline looks good to me.
To make the import and processing faster and much lighter in terms of memory, you could avoid importing the entire file as a single epoch in the database, by using the menu "Review raw file" instead of "Import MEG/EEG".
https://neuroimage.usc.edu/brainstorm/Tutorials/ChannelFile#Review_vs_Import
Your problem could be due to several things:
- a reference issue (Source Localization Abnormality - EEG reference problem)
- a problem of electrodes positions (have you checked that the electrodes are nicely located on the head? - please post a screen capture that shows the electrodes at the surface of the head)
- a problem in the OpenMEEG computation: try using a spherical model instead
- a failure specific to sLORETA: try using dSPM instead, or a non-normalized map ("Current density map")
- a problem due to the absence of noise covariance information: compute a noise covariance from a short segment of recording where no major event occur
I don't know if any preprocessing step performed by EEGLAB could cause additional trouble...
Have you tried reading and processing the continuous files in Brainstorm?