This is a known issue; without any further details or a way to reproduce it, this can't be fixed in the plugin, in Jenkins or by the Android tool developers.

https://wiki.jenkins-ci.org/display/JENKINS/Android+Emulator+Plugin#AndroidEmulatorPlugin-AVDsmaynotshutdownfullyattheendofabuild

One workaround is to periodically run a cronjob that forcibly kills any emulator processes which have been running for a certain period of time (in this case 10 minutes or longer):

ps -U jenkins -o pid,etime,comm \
  | egrep ' emulator$' \
  | egrep -v ' 0[0-9]:[0-9]{2} ' \
  | awk '{print $1}' \
  | xargs kill -9 2> /dev/null

Used this workaround for some time, but it isn't very flexible when working with different project. Some project can run tests longer then 10 min. Some fits in small time interval and emulator needs to be used faster then 10 min mark.

Did anyone even tried to run Jenkins in any Linux dist with such config:
*Android plugin
*In one node at least 4 executors
*Multi-configuration project with running instances parallelly.

Noticed that there are less unclosed emulator instances when less emulators are running. Basically running 1 project with 1 emulator instance never seen unsuccessful closing.

Yes, I use that configuration every day, with multiple nodes, each with four executors running in parallel.
We also have some long-running jobs and, while the emulators don't always stop, I use the above "kill" workaround to stop emulators that have been running for longer than two hours (since we have some jobs that run for three minutes or so, and longer ones of 90+ minutes).

I have also tested the plugin on a machine with at least 12 parallel builds (mentioned at the below link), and there were no issues with the emulators shutting down improperly.
https://issues.jenkins-ci.org/browse/JENKINS-7354?focusedCommentId=142210&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-142210

I don't have time to try myself, but does this code works as it should?
AndroidEmulator.java:525

boolean killed = sendEmulatorCommand(launcher, logger, userPort, "kill");

// Ensure the process is dead
if (!killed && emulatorProcess.isAlive()) {
   emulatorProcess.kill();
}

In this situation emulator thinks that it will be closed properly so killed == true so emulatorProcess.kill(); is ignored...

Originally, only the emulatorProcess.kill() command was used when trying to kill the emulator process.
Because that didn't always work, the sendEmulatorCommand was used instead in an attempt to make shutdown more reliable.

So I don't believe unconditionally calling kill() here would add anything.

I have been seeing a similar issue with the emulator not shutting down properly. So far I have not isolated a specific test case that reliably reproduces the issue. I have however spotted a number of issues that may contribute to the problem. I will log these in separate JIRA to keep things simple.

Marking old "resolved" issues as "closed".