Hmmm, this would not be my first choice.
Express MaxDiff works great for sample-level utilities, but we know that the respondent-level correlations between Express MaxDiff utilities and the utilities that would come from a standard MaxDiff aren't all that highly correlated. So I could imagine the segments might differ quite a bit from what you would have found with a more valid set of respondent-level utilities.
That said, if all you have are Express MaxDiff utilities and you want to do utility-based segmentation, then you don't have a lot of options do you? As long as you can live with the caveat that the utilities aren't ideal for segmentation, I'm sure you could give it a try.