I wouldn't call it a myth, I'd be more inclined to call it Rare Problem. From what I remember, problems seem to usually happen on very busy pages with a LOT of content for the browser to parse through. With the browser computing very quickly sometimes a lone inline-element (between blocks) lower down in the page source can get misplaced. It's not easy to reproduce the problem in a simple stand alone test case.
Your right, it's not in the specs, in regards to being illegal. But here's what is in the specs which I would say is VERY closely related.
220.127.116.11 Anonymous block boxes
In a document like this:
(and assuming the DIV and the P both have 'display: block'), the DIV appears to have both inline content and block content. To make it easier to define the formatting, we assume that there is an anonymous block box around "Some text".
In other words: if a block container box (such as that generated for the DIV above) has a block-level box inside it (such as the P above), then we force it to have only block-level boxes inside it.
You see, the browser is being forced to generate an anonymous block box around any inline elements that are siblings of a block within a block. If you don't wrap your inline elements in a block yourself then you are depending on the browser to figure it out on it's own. That inline content is going to get a block box around it one way or another. By doing it yourself you take the burden off the browser and allow it to parse through the page without having to stop for a millisecond and make a block box.
As I mentioned above, we usually see the problem crop up on busy pages. When people post threads about it, this comment is almost always found in their post... "When I refresh the page everything is fine".
I'm guessing that's because the browser got the anonymous block box in place the second time around.