Calibre creates a million css tags with lit files in the current version. Of course that makes trying reformat the file by modifying the stylesheet nearly impossible. .6 css creation has been completely re-worked to resolve this, but you'll need to wait a few weeks until the beta. It looks like lit2oeb is creating better output than Calibre today, so that's a better starting point. You want to stay away from amber lit converter.
Regarding line breaks in all the wrong places, that's because many lit files are pirated, and they're straight OCRs - the original scanner just left all the lit file line breaks where they were in the original printed book. There is a new feature in .6 that will allow optional post-processing during conversion to resolve issues like this - the framework is there, but lit files don't have the regex replacements in place yet. .6 is still several weeks away so it will probably be in by then.
See this bug for info on the new function in case you're interested in contributing:
http://calibre.kovidgoyal.net/ticket/2359