Unfortunately, the HTML is not mine and I do not control it.
I took a closer look at the View->Source and indeed it is malformed with respect to several of the tags. The <i> is not matched with a corresponding </i> and some of the <tr> nodes aren't properly closed by a corresponding </tr>.
Nevertheless, I have several observations:
1) The AOT DOM is correct when the browser is FireFox - the malformed <tr> is correctly set up as the child of the <tbody> even though the </tr> is physically missing from the original HTML. In other words, FireFox does the parsing properly even though the original HTML is malformed.
2) The IE DOM, as displayed by the Web Developer BHO plugin, is also correct - the <tr> was elevated to be a child of the <tbody>. In other words, IE also gets the DOM structure right even through the original HTML is malformed.
3) If you look carefully at where the first malformation of the <tr> occurs, you see that it happens after the AOT DOM error of the previous <td:2>. I would expect the malformation of the <tr> to affect subsequent nodes, not prior nodes. It does not make sense that a later error in the <tr:2> should have caused a <td:2> which was properly nested (in the HTML) to a <tr:1> to become improperly nested and become a child of the <tbody:0>.
4) The AOT DOM parser handled the missing </i> nodes properly. It recognized that the closing </b> node must be implicitly closing the <i> node. I do see that this is a different type of implicit closing of a tag. In the <i> case the presence of a '</' on the </b:0> is clear indication of implicit closing, but in the <tr> case, there is no closing </tr> to trigger a simple implicit closing. In this case it is triggered by 'global' knowledge that <tr> nodes need to be children of <tbody> nodes.
Suggestion: Improve the AOT IE DOM parser to better recognize implicit tag closing. The parser should be smarter and recognize that <tr> nodes are only valid as children of the <tbody> hence when a <tr> shows up at the wrong nesting level, promote to be the direct child of the <tbody> and thus implicitly close all the open tags.
Also, in IE's case, perhaps your native DOM parser can 'consult' the real IE DOM to discover when it parses differently from IE's parser.
-David