Hi, what you could try is to make sure that all your seeds are really binary (i.e. only 0 and 1 in the images). Then it will be much faster and should also not consume as much memory. The lead mapper tool can use weighted seeds and these will be processed more elaborately given tracks need to be weighted for each voxel.
So if you break down a parcellation into single files, make sure to use nearest neighbor interpolation or do it manually making sure your nifti file only contains booleans (e.g. no values like 0.9999 or 0.000001 at bordering regions).
Other than that no clue how to fix it easily.