James,
I assume that you do have more than one trial for each movie type since you're using dmUBLOCK. If so, the results from TENT should in general give you more accurate characterization for the BOLD response compared to the one from dmUBLOCK. The reason dmUBLOCK assumes an idealized response curve with a prefixed shape while TENT estimates the shape (and the amplitudes) from the data. In other words, I would definitely trust more the result from TENT in terms of response characterization. I would not worry too much even if the two results cannot resolve with each other.
The impact of unmodelled part with TENT would depend on a couple of factors. Were the movie clips randomized in terms of their sequence as well as the time intervals between them? How many repetitions do you have for each movie type? If the experimental design is fine, the impact of those unmodelled time points would be mainly on the residuals and statistics, but less on the effect estimates.
Gang