It's hard to see the problem without images. Could you attach a jpeg image? The @AddEdge script will help you look at the results, and an image from there would be useful for me to help. The align_epi_anat.py script assumes the datasets start off fairly close, within a few mm of each other. If the distance is more, then you can use the "-giant_move" option.
Also the epi_base can be set to a volume that shows greater contrast (distinguishing CSF from white and gray matter). A volume before reaching steady state (-epi_base 0) may work out better.
I don't understand what you mean by "the actual epi image is proportional to the tlrc image." I am hoping you are comparing the right kinds of results.