I'm more than a little behind in posting.
Every year around winter I get into an art mood, and work on either ambigrams or origami architecture cards. I've done a few ambigrams before, but I didn't think about the psychology behind it until now.
When I first started playing around with amibgrams, I showed a few to other people. Someone then commented that it's just a visual trick, because humans tend to fit what they see into categories they are already familiar with. In this case, we only see letters the right side up, and ignore the other potential interpretation of the letters being up side down.
Thinking about that a few days ago, I was reminded of how humans seem to have a special brain area for recognizing faces. Humans apparently need to recognize faces so frequently that by two months there is already a brain area which activates on perceiving a face. We have a tendency to see faces in random objects, and some people have a selective disability in recognizing faces. More related to ambigrams though is the Thatcher effect, how humans can't immediately identify "problems" in upside down faces, which are otherwise rather... eye-catching when viewed the right side up.
I think the same thing happens with ambigrams. We become so accustomed to reading letters and words right side up that we ignore small features changes. It just so happens that these changes form the same letters (or sometimes different letters) up side down, hence creating the ambigram.
If this theory works out, words that are more common would be easier to see, because of the reader's familiarity with that sequence of letters. For example, someone named Earnest might be faster to read his name in an ambigram, compared with other people who would see the adjective. Seems like it would be a pretty cool psychology experiment.