It's not just guessing. The Carpalx does have algorithms for scoring various characteristics like finger bigrams, finger travel, and outputs the optimal results based on scores for key presses and bigrams, etc.
The metric I lean on the most is the one published by Andong at http://www.andong.co.uk/dvorak/Default.aspx. I guess that's odd since I spent the least amount of time talking about it and put it at the bottom of the page. It is focused on metrics of efficiency and effort. If you wanted speed, you might want to look at things like finger distance and home row usage more closely, in which case Arensito or Colemak might have an edge.
I must confess, since I swapped R and H in revision 2, Workman usually gets the edge on Norman using Workman metrics, depending on the input text. I guess I'll hunt down whatever text revisions are appropriate on normanlayout.info.