Everything is much simplified if you just avoid this weaving altogether, and adopt a content pipeline approach, as every commercial book publisher, newspaper and magazine does. In this approach, you use pure-composition programs for your composition, and pure-layout programs for your layout. When you are working on one, you are not distracted with the demands of the other.
In fact, two different people can be working asynchronously on these two components; and, indeed, if the work of one stage is finished first, this enables an extremely useful static constraint on the work of the other, simplifying the decision-making for it greatly. Layout can determine target word-count; or word-count and paragraph composition can decide layout.
What you don't have is the constant tweaking, pitting half-finished layout and half-finished prose against one-another. Each is effectively opaque to the other until the end of the process.
The most wonderful thing that has happened to composition in recent memory is the GUI Markdown editor. Until these, all composition programs that weren't simply plaintext, had at least some layout elements. (This is because their target format was RTF, and RTF contains things like per-line margin settings.) But .md is a fine format for storing compositions: it lets you specify the "tone" of text with markup, but contains no mechanism to (reliably) specify final appearance.
Combine such documents with a program like Publisher or InDesign (or, partially, LyX or Pages) for layout, and you've got a very nice workflow. And, as a side-benefit, the prose created during the composition step will be exactly what is needed for web publishing as well.