I've found that when I plot the condition means for these two models side by side (with trial collapsed, so they can be compared directly), most of the mean estimates are fairly similar. Because of this, I am guessing there is nothing wrong with the way I used 3dLSS, but rather that the different ways of modelling the data caused relatively small differences in how the means were estimated, which resulted in noticeably different ANOVA outcomes. So, I guess my take away is that statistics are finicky.
