I checked the SMU service manuals I had in front of me, Keithley 237 and Keithley 2400, and both take it after the current sensing. The 237 from the current sensing amplifier and the 2400 via the sense line where ever sense is connected (at the binding posts on the back or at the load). See attached block diagrams. I'm sure this (together with the current sensing itself) will introduce some bias current that affects the current measurement, but as long as the bias current is stable, this can be corrected during adjustment. I think your explanation that there could be a substantial voltage drop over the sense resistors is the most obvious one.
I would consider asking a moderator to move your topic to the test equipment section (use the "report to moderator" link).