» If I understand it correctly, Tschichold recommentds only old-style
» digits for text (he calls them text digits as different from titling
» digits). Therefore I think that in the tradition of old school there
» should be difference between '3's in the phrase
»         In Chapter~3 we will show that $\pi>3$.


Nope. The really old tradition did know nothing about lining (also
called "english" in France ;-) digits, anything used old style, which
you still can see in the scan of a rather recent book by H. Cartan
I posted somewhere in page of the Latex navigator.

There have been modern attempts to do what you're refering to. (I had
a vote about the 3 possible styles, but I cannot make statistics about
the voter's taste, as they were so few -- you can have a look at
http://www-fourier.ujf-grenoble.fr/~bouche/tex/mathtests/OS-ornot-OS-e.html
if you have some spare time to loose...)

» Also, the command \MakeUppercase should take care of this
» distinction, because old-style digits are actually *lowercase*, and
» must be converted to upper case when, e.g. in running heads.

yep. this is easily done by adding a family switch to
\MakeUppercase. One more reason for a `case' axis in NFSS, but one
more complexity (you would need families with varying letter-cases and
fixed figures-cases + all other possibilities round...)

» What do other people think of this?

I think that you're point is one point in a myriad of
possibilities. If you replace `must' by `could' in your discussion, I
heartily agree with you, but notice that what you describe is easily
done with current latex (redefine most \the, \MakeUpperLower-case,
etc.)

Thierry Bouche
__
  « Ils vivent pour vivre, et nous, hélas ! nous vivons pour savoir. »
                                Charles Baudelaire, Paris.