MediaLab *does* do reaction times but DirectRT is better when extreme precision is required. For questions like what you are describing, MediaLab response times will usually do the trick. Any error in ML's RTs is usually offset by variance in the reading time of a question like "how much do you like ice cream" and then the subsequent selection of a scale response. This would be in contrast to "hit this key if you see a real word and this key if it's a non-word" sort of thing. For more on this see also:
http://www.empirisoft.com/support/showthread.php?t=15