I assume that your trials are event-related. If that is the case, how to model reaction time is a moot issue. There are two straightforward approaches, each of which carries slightly different assumption and provides you different results and interpreations:
1) Making a crude assumption of linearity between the BOLD response and RT. This is typically done in amplitude (or parametric) modulation with -stim_times_AM2. With this approach you get the response estimate that corresponds to the AVERAGE RT (plus the modulation effect).
2) Treating the RT as the duration of each trial. You can model each trial with a duration (e.g., dmBLOCK) of the RT value with the assumption that RT acts in a proxy role of stimulus duration. In the end you obtain one response estimate that is associated with the unit (not average) RT.
It would be interesting for you to see how the two approaches pan out. And the choice would have impact on the interpretation of the final results as well as your group analysis.
Gang