a/b testing is a great tool, but wouldn't telling people they are participating in an a/b test trying to determine a specific question 'requesting e-mail addresses from visitors' bias the results?
310 unique visits / 22 addresses captured (7%)
50% of traffic from news.yc - and generally didn't leave an email address at all.
Users from other sources were 10 times more likely to leave an addresses if shown the pop up on page load.