I agree the proof is in the pudding and benchmarking is crucial to having valid opinions but I've seen the pudding served to 10-20 million users a day all year and even minor changes like moving a linked heading lower or higher relative to other page blocks does change usage. This isn't really an experimental problem any more. What if someone moved your mousepad every morning? Oh, that's amusing.

As discussed, a pure design site or something might make constant change interesting or desirable, otherwise it's probably a vanity choice and not really in your users' good interests. And even the design sites that do this kind of thing stay strongly branded with a big headline/logo/etc. I can't think of a single site I've visited more than once in 10 years that does this though. On the other hand, letting users select a template or stylesheet is most definitely in their interest and good service. And if it's your site, it's your call.

The other problem with doing it via two different code bases instead of CSS or templates is you just locked yourself into a future course if you intend to keep users who do like the different systems happy. If you upgrade, redraft, change then your costs/time just doubled. If you can change one of the systems to run with the template of the other (the end product of HTML doesn't care what created it) you'll probably be happier as time goes by. Plus you're in an unusual and enviable position. You have two code bases done and running, and you could pick the better written of them.