Dropping in here to say that we've certainly been aware of comments along these lines.
Unfortunately it is very difficult for us to answer every thread on the forums, or to debug issues that we aren't able to reproduce ourselves. The forums are primarily intended for peer-to-peer API discussions and direct announcements, although some of us try to be as responsive as we can to the different categories here. We're not able to be here 24x7.
Personal speculation - there's a chance that some backend optimisations may have helped; that a Mesos host may have changed and cached / invalid paths were being hit; or that something outside of Twitter's network could have affected this.
Glad to read that some folks are seeing better performance, but as with many large-scale web systems, unfortunately there's a chance of occasional 404s along some of the API paths.