I don't see why people shouldn't expect passport Ryo... after all just a few months ago they photoshoped Ryo's DC face into a screenshot, implying they were still going for that look.
Ryo's in-game model is not passport Ryo, though.
As you can see in the video I posted, they literally introduced Ryo for the first time by actually morphing from a Miyawaki design to the in-game model. That's arguably always been their approach to the characters in-game. Passport Ryo is a Dreamcast technical showcase.
The morphing doesn't prove that the design went straight from Miyawakis 2D-drawings to the in-game characters and it clearly didn't. Ofcourse his drawings are important as the first version of a character.
The point is that the in-game models were not meant to replicate those drawings. As Yokosuka pointed out,
the 3D CGI images were the targets for the in-game models. In-game Ryo is a low-poly version of this:
https://static.giantbomb.com/uploads/square_small/3/34651/1188145-0ryo3.jpg
and the passport-models are simply more detailed models with the same target.
It's pretty clear that the design process for the main characters was something like this:
Miyawaki makes concept drawings of a character, then a 3D/CGI-model is made based on this concept but much more realistic, detailed, and different from Miyawakis "comic-book" style, although still stylized. For some characters clay busts are also made.
Then the actual model used in the game is made to look like the CGI image but they obviously can't fully replicate it in real time.
So, to take Joy as an example:
Original concept drawing:
http://www.hardcoregaming101.net/shenmue/concept-joy.jpg
CGI model:
http://i.neoseeker.com/ca/shenmue2_conceptart_njFRi.jpg
In-game model:
http://cdn.staticneo.com/w/shenmue/thumb/4/4e/Joy.jpg/300px-Joy.jpg
Thanks to great modeling and texturing, the in-game model manages to capture the essence of the CGI image although less detailed.