In the olden days, A/B testing of newspaper adverts took advantage of a printing system in which two otherwise identical copies of a newspaper was printed on wide rolls of paper. This fixed the testing ratio at 1:1. For new website features, this doesn't have to be the case.<p>Pick a small sample of random users, perhaps where account creation time modulo 1000 equals 120. Make new code paths available to these users only. Measure load. Get feedback. Then repeat where the userbase is larger. Repeat until the feature is available to everyone.