Yurie, thanks for reporting the issue. Can you run sh "env" in your wrap block? The DISPLAY environment variable should be set there. Also can you provide the versions of the Xvfb plugin, Jekins and console log of your pipeline job?
I believe I've tracked down the cause of the problem, in fact it was two issues that were causing this behavior: race condition between the test if Xvfb was running when started by docker exec and use of frame buffer directory for Xvfb that was not mounted as a volume in the started docker container. Now the timeout value defaults to 1 second, that should be enough to alleviate the race condition and frame buffer directory is created within the workspace of the build that should be present in the docker container.
I've released a beta version of the Xvfb plugin to the experimental update center could you install the version 1.1.4-beta1 from there and test if this fixed your problem?