This is really hard because it is like making a guess based on only having 10% of the equation
First of all you are saying your conversions drop... is that for every single keyword? You are tracking at a keyword level right?
Generally the more specific the keyword, the more specific you should have a landing page
You should only be testing with 20% of your ads... thus you would have A A A A B - if you are switching things around wholesale, you have no data to really work with - even external factors like the world cup can make a difference with sales.
Play with keywords you know convert first because they give you possibly the biggest volume and you can work to improve. Create multiple versions of a landing page for your top converting products and split test between them, but also perform multivariate tests on differnet layout elements.
If there is some way you can create the landing page such that you are not constrained by the shopping cart, even just for one item, thus you might grab the html from an existing page then modify it by hand without theme limitations for the whole site.
I have no idea what might have been happening with your quality score through all these tests, but a different quality score could affect your ad position or bid price on just a few keywords... and those happened to be the ones that bring in the conversions.
