%First, making mathsurround a glue:
%So as not having to type something like this, which could be solved better by
%\mathsurround=1.11pt plus 1.66pt minus 0.5pt, say.
%
%This is some text at the top of a page with a figure at the top-left corner occupying more than half the with of the page,
%so the text has to be typeset in a narrow column. A similar case would happend for publications that display their text
%in serveral columns. (The \- was not needed in the original text, of course; just for right hyphenation)
%Definitions for some macros so that the example can be run.
%Some of them have been crudely defined, but for the example is enough
\def\sdot{\mskip 2mu minus0.6667mu\mathchar"0201\mskip 2mu minus0.6667mu}
\def\diez#1#2{#1\sdot 10^{#2}}
\def\?{\mskip -2mu}
\def\punto{.}
\def\ApT{A^{\prime T}}
\def\sloppy{\tolerance=9999 \emergencystretch3em}
%End of macros
\font\ften=pplr8t at 10pt \ften
\frenchspacing
%beginning of example
\quad\vbox{\hsize 114pt \sloppy \def\s{\spacefactor2000}
$$
\hbox{[Omitted text]}
$$
y para\s{} $\Sigma_A^{1,2,3}$\s{} y\spacefactor3000\penalty10000{} $\Sigma_E^{4,5,6}$\s{} es $\diez{16}{-6}/\?\sqrt{3}=\diez{7}{-6}$. Se otienen así $A'$ y~$L'$. La matriz $N$ es~$\ApT A'$, y $N^{-1}$ se muestra en la página si\-guiente.\par
}
\end
%Appart form these situations, it is customary to consider that larger spaces shoud have larger total strechability, both in horizontal
%an vertical lists, so I don't see why TeX should ban that possibility for the mathsurround space. Consider the paragraph at the bottom
%of p. 72 of the TeXbook:
%
% [...] furthermore, it automatically increases the strechability (and decreases
% the shrinkability) after punctuation marks. The reason is that it's usually better
% to put more space after punctuation than between two ordinary words,
% when spreading a line out to reach the desired margins.
%
%Why should this argument be right for the extra space after punctuations but not for the extra space arround math?
%Now the second proposal: a primitive that inhibits the insertion of the mathsurround space when there is no space
%at all between the math and the following (or preceeding) character. Well, it seems obvious to me.
%I once tried to typsed a math problem with a nonzero mathsurround and the space that got inserted before the commas
%and periods looked so awfull that I desisted.
%
%What I proposed for the precise definition of "space following" is that at the time of packaging the list there be
%a glue node following the end-of-math node. If there is none, the space is not inserted. This is very easy to implement
%since the amount of space is carried by the math node, and not inserted itself as a node.