Thanks for your answer, but I am still unsure on the best way to proceed.

Let me clarify a bit:

I have two days of data for each subject, 9 runs on a given day.

For purposes of a conjunction style analysis, I want to ultimately analyze the data in a common space for each person--Day 1 functional space.
This will involve registering Day2 runs to Day1 runs.

If I normalize each Day2 run, then move them to Day1 space, concatenate them, then run 3dDeconvolve, the resulting beta values will still reflect % signal change, right? In other words, does the interpolation of the data involved in the registration affect the values in a such a way that they are no longer "normalized" according to specifications?

If necessary, I suppose, I can normalize after registration, but because I have 9 runs of data and would prefer to move them all in one lump, I would rather normalize before registration.