When a connection to a slave is unexpectedly broken during a build, an Xvfb process started by the plugin is not killed. After the slave is reconnected, subsequent builds on it fail immediately because the plugin tries to start a new Xvfb process on the same display as the old one.
It would be helpful if the plugin could kill left over Xvfb processes before trying to start a new one. There could be a configuration option to kill all Xvfb processes. More robustly, it could probably find just the problematic one by command line.