In the early '90s when I was a usability specialist at Microsoft I did a laborious usability study (hired several temps to enter text using various methods for hours per day, over multiple days) comparing pen input vs. OSKs and got some counter-intuitive results. Essentially OSKs are much faster than handwriting even under optimal conditions. If it takes 10 seconds to enter a string with an OSK, it'll take ~15 seconds to write it using block letters. That's not counting any time needed to correct errors, which will tend to make the writing input even worse. When I asked participants after the study which method they thought was quicker, most assumed the pen-based writing was. Keep in mind that they spent 1/3 or more of their time correcting errors, so the writing may have taken twice as long as the OSK, yet they didn't perceive as such. Of course their statements could have been a demand effect, but it generally corroborates my main point.
I've read a pile of published usability reports on other text entry methods, such as chording keyboards, and they all show essentially the same thing. Letting fingers extend to hit a fixed close target is hard to beat, and the optimal amount of pressure needed is very small (though perhaps not zero).
My theory is that it takes less mental effort to write than to use an OSK, so the OSK is perceived as being slower. With the OSK the level of mental effort is mostly high (mostly visual search), with periodic dips (hit the key). With pen-based writing it's the opposite -- mostly automated / low-cognitive-effort behavior with periodic thinking about what to do next.
I read a quote from Jobs somewhere recently where he said something like, "the problem with handwriting is that it's just so darned slow" so that makes me think that Apple may have done a similar usability study.
I don't think OSKs are ideal. I'm continually impressed by the steady flow of innovations in text-input methods. But pens have their own usability baggage even outside the limitations of handwriting.
