I concur with Joe and David. The text from 8879, 9.4.5, says:
"The refc or RE can be ommitted only if the reference is not followed
by a character that could occur in the reference, or by a charracter
that could be interpreted as the omitted reference end."
The general problem here is that you are attempting to paraphrase 8879
as you do in other places in the text. This is the dangerous in that it
may introduce alternate interpretations of the lexical behavior as required
to remain SGML conformant.
I'd go further and forget trying to be nitpickingly accurate wrt The
Std., and just say that you've got to end entity references with a
semicolon unless they are immediately followed by another one
(beginning with an ampersand).
SGML gureaux know perfectly well that this is not the case...and for
them it doesn't matter. Software implementors should probably be
reading more than just the HTML spec if they want to implement serious
software, and it's sufficient for the average user to make sure they
don't fall into the trap of writing AT&T etc.
///Peter