-lpc divides the image into a bunch of sub-volumes, computes the weighted correlation coefficient of the image pair over each sub-volume, then averages a function of that collection of correlation coefficients (giving higher weight to the larger correlations, both positive and negative). The goal is to make the most negative correlation on a local basis (default radius of the sub-volumes is about 6 mm), since EPI and T1 images are approximately inverted in contrast in most regions -- but not in all.

The -lss cost function is just the global (weighted) correlation -- the goal is to make this correlation as negative as possible. It also works pretty well for EPI-to-T1, but seems to be less robust than -lpc.

Recall that this stuff is preliminary, and subject to changes, to caveats, and so forth. In particular, -lpc was implemented only 2 days ago, and compiled into the binaries on our Web site last night. That's why using it is still rough around the edge -- the bleeding edge.