Maybe to add to the previous answer because I think it does not adress the core question:
I think there are two reasons you might want to jitter your stimulus onset in relation to the TR. 1) Design/shape considerations 2) psychological considerations
1) Design effect: If you are only interested in BOLD amplitude, then you might not want to jitter. This allows you to do very simple analyses without any deconvolution/hemodynamic basis functions. Because you measure the BOLD response at the same point in time, you can simply average the 1st,2nd,3rd etc. TR after stimulus onset.
But if you are interested in the shape of the BOLD response, then you might miss the peak, miss the undershoot etc. Thus for an acurate estimation of the shape, you should jitter the stimulus onset. In other words, jittering gives you a higher sampling rate at the cost of less samples per sampling point (less power without assuming a BOLD shape).
In this image you can see that on the left side, by chance (or on purpose) we sample very close to the BOLD peak - great! On the right side, we jittered the stimulus relative to the TR. We now
sample the BOLD response more uniformly, but averaging is more difficult.
Note that this is related but a bit orthogonal to the event-related / blocked design decision.
2) Psychological Effect: If you always present the stimulus at integer TRs (e.g. with a TR of 2s), subjects will synchronize with your experiment and expect the stimulus. Whether this is something you want, you need to decide for yourself.
More information can be found here: http://mindhive.mit.edu/book/export/html/74