I'd also be interested in DM or "duration modulation", where the duration of each event is modulated by another variable. FSL allows this in their "3 column" format (onset, amplitude, duration).
A recent paper by Grinband et al.* shows that amplitude modulation doesn't do that good a job of modeling variation in the HRF due to temporal factors even, apparently, for short time scales (e.g. reaction time). AM seems to be better suited for changes is stimulus intensity rather than changes in RT.
* Detection of time-varying signals in event-related fMRI designs.
Grinband J, Wager TD, Lindquist M, Ferrera VP, Hirsch J. 2008