You say

“(…) will therefore adversely ef­fect a lot of users (…) Especially dis­ap­point­ing in IE (…) It’s upsetting to see this, really. (…) now it seems we have to ei­ther shy away from them or simply accept the side-effect. (…) This is breaking behavior that greatly impacts readability (…) it can be surprisingly noticeable and very distracting.
“

I think you over-excessively exaggerate the damage (severity, gravity, criticality) caused by this bug. It’s 1px. And I am convinced that a very wide majority (over 99.99%) of ordinary people will never notice such a bug. Your demo page is also very revealing of this. You need XP and a good, careful, thorough look to see the visual effect of the bug: a 1px offset.

On the other hand, websites with tiny font size and frozen font-size tyrannizing users, impacting readability, especially for seniors, is still very frequent.

I also checked the code of your webpages, including the code of the test page that reproduces the problem. You should first fix the validation (markup and CSS) errors. e.g.
line-height:22px; !important;

Your CSS code is unjustifiably over-declaring, also over-constraining.
When trying to “corner” a bug in CSS or in HTML, you need to reduce code declarations. Your testpage does the reverse: it over-declares. E.g.:

font-family:'cambria';
font-size:17px;
line-height:22px; !important;
Why setting font, font-size, line-height? If the bug is really caused by the entities spacers, then you should avoid including lots of CSS declarations; you should reduce CSS declarations to eliminate possible side effects.
Why did you want to set !important to line-height anyway?
You expect to get 2.5px (22px minus 17px divided by 2) for half-leading above and below content area. It does not work like that. Fraction of a pixel is not rendered the same in mainstream browsers (IE8, Firefox 3.6, Opera 10.5x, Safari 4.0.5, Chrome 4.x, Konqueror 4.4.2); some truncate, some round up.
any standard web font will exhibit the behavior
Your testpage does not demonstate that.

position:absolute; left:40px; etc.
Wasn’t it possible to create a reduced testpage without having to absolutely position text content? Not sure…

text-align:justify;
text-align:right;
text-align:justify;
text-align:left;

Why the double text-align declarations?

“The jog can also be more easily seen if you Zoom up to 125% or 150%.”

Well, then I think it would have been a good idea to set font-size: 2em; (or font-size: 200%;) and then write one, only one, 2-lines sentence and then put both of them side by side. Also I would have tried to reduce column width… instead of fixing them to 444px.

Maybe there is a bug with the line-height… but your testpage does not demonstrate this well and correctly. There is just too many issues with your testpage. One good point is that you provided 2 screenshots…

Regarding Zoffix Znet’s comment, overall, I agree entirely with Zoffix Znet. Web design isn’t print design. Web authors should focus on content and less on style and presentation; users should have control over presentation. Users have a veto power over submitted style and presentation in webpages.

regards, Gérard

And sorry Zoffix
“if you want to control user experience to the pixel then you picked a wrong profession.” lol…. funniest thing I’ve heard all week, you probably still use tables Zoffix with 4px borders.

I was reading your observations about font rendering, as well. I’ll be going back to take a closer look. Good stuff. Bookmarked!
Yes, you’re correct. The CSS term ‘line height’ doesn’t map exactly to ‘leading’ in the print world. thanks for bringing up the distinction because I was planning, at some point, to write a glossary of print-to-web terms.
Regards, Rich

Second of all, I could not reproduce the bug when using text-decoration: underline overline; and line-height set in pixels (and it shouldn’t be set in them) to make those two lines touch each other. Perfect in IE8 on Windows.

Lastly, if you set line-height in `em` units, the line height vary by a pixel in Firefox and I’m sure in many other browsers.

Cheers!
Zoffix Znet

You’re so 2009 you’re 1997!