This was originally going to be a comment on Joe's recent post, but it started to get too long, so I posted it here.
As usual, I'm about a month late in commenting on this, but if anyone's still listening, this really is an interesting phenomenon. However, I have not yet been able to find the actual study from "Cmabrigde Uinervtisy." (It seems to be loosely based on an unpublished 1976 PhD thesis, but I'm not sure about that.) Although this is certainly a powerful testament to the versatility of the human brain, it seems that another (probably more important) factor in this is the fact that the English language generally uses short words.
An average paragraph is mostly 2 and 3 letter words -- these are not mixed up at all by this algorithm. Words with 4 letters have only the middle two switched. (Note: It appears that the average word length in English is about 5.1 letters. Sources here and here.) Once the short words are determined, the others can be much more easily determined by context clues. If you were to have a paragraph with only words of 6 or more letters scrambled in this manner, it would be much more unreadable. Similarly, if you were to scramble letters not fixing the first and last, a regular paragraph would be much more difficult to read.
I saw a similar post which demonstrated how difficult it would be to read a sentence like the following:
Setl yrt an txperimene. Nac uoy dear eht text eerh?
(Don't read on until you think about it.)
In the preceding sentence, only the first and last letters were scrambled. So, clearly the first and last letters make a difference. However, you can also see how important context is. Try looking at the following words:
Now try it with context:
Tihs llitte bit of iuigrnnitg lgutisinic trviia srtmeod turhogh inxebos in Seempbter 2003.
This little bit of intriguing linguistic trivia stormed through inboxes in September 2003.
Also, if we look at a paragraph of almost mostly long words:
Pahprgaras cseopomd uisng wrdos cintonnaig laregr nburmes of ltetres cusae nuumeors dutecfiifils for radeers sekenig undsrdennatig trrefeohm when lrettes bcoeme sbarlcmed, eevn wehn mainanntiig the pnosioits of iianitl and fnail cratcehras.
This is much more difficult to read. (Of course, such paragraphs are naturally somewhat more difficult to read, even when left unscrambled.)
Also, even a relatively simple paragraph can be very difficult to read if we scramble the letters and don't fix the first and last letters:
Is hte eordr fo het seerttl in a wdro lelary autipnnortm? sI ti urte atth ew ilsypm see eht rdwo sa a hleow? Whta seel oteburcitns ot sith onnonhmpee?
With first and last fixed:
Is the order of the ltertes in a word relaly upnoinmtart? Is it ture taht we simlpy see the wrod as a wlohe? Waht esle ctneobrtuis to this phennomeon?
Is the order of the letters in a word really unimportant? Is it true that we simply see the word as a whole? What else contributes to this phenomenon?
Nonetheless, this is a great post and a very interesting point. (Related links here, here, here, here, here, here, and here.)