The idea would be that normalization/standardization can be applied a posteriori, for instance after several runs are averaged at the source level. dSPM and sLORETA can also be expressed as linear kernel operators, but the issue is with the proper estimation of data covariance on single trial data: not impossible but not entirely convenient either.
So in summary, we do recommend that a form of standardization be applied on source maps: this latter can be performed at different instances of the source mapping pipeline.