If there is problem with SCM polling hanging (as in JENKINS-4461) and you try to manually start a job, Hudson will duly schedule it on an executor and then seem to run it. When you look at the console, it says e.g.

--%<--
Console Output
--------------
Started by user jglick
Building remotely on someslave
--%<--

and then nothing else happens - indefinitely. This is not friendly.

If the build has to wait to acquire a workspace lock, it should say so... and give you information about how to kill any outstanding polling process (JENKINS-4479).