Thanks very much for your comments and your interest in MathML.
I am leaving for a week of traveling tomorrow, so I wanted to respond
before I go. However, I'll have to be brief. Perhaps someone else
from the working group will pick up the thread.
Our intention is that the only legitimate places to break a line is
between items in an MROW. We will try to say more on this. Other
people have commented that the name MROW is misleading, and I would
welcome suggestion for a better one.
On stretchy operators, etc.
As you point out, there is no automatic mechanism for doing "global
optimization" of the typesetting in expressions, for example to make
integral signs the same size. However, it is still possible to do
this by hand, using "minsize" and "maxsize".
Another point worth making is that math typesetting style handbooks
such as the one you refer too (I am not familiar with it -- thanks for
the reference) were written in an era of metal type. At least one
working group member contends that there is no compelling reason to
limit things like delimiters to discrete size increments on a device
capable of displaying any size. I myself would like to stick
relatively close to traditional typesetting, but I do take the point
that good typesetting on paper and good typesetting on a low
resolution screen might be different.
This is a real issue, also raised by others. It must be clarified.
If you have ideas on the subject, I would welcome them.
HTML Math working group co-chair