Created attachment 26302 [details]
Deadlock thread dump.

filesystems aren't involved at all -> lookup versus datasystems.

I wrote a test, but alas, it seems that the deadlock in 67472 cannot be 
properly fixed in XMLDataObject, that would create possible access to 
not-yet-initialized Processor object. Leaving for htmlserver guys to fix. 
 
Checking in XMLDataObjectTest.java; 
/cvs/openide/loaders/test/unit/src/org/openide/loaders/XMLDataObjectTest.java,v  
<--  XMLDataObjectTest.java 
new revision: 1.3; previous revision: 1.2 
 
The actual problem is the initialization of   
  
org.netbeans.modules.httpserver.HttpServerURLMapper.<clinit>(HttpServerURLMapper.java:37)  
at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)  
  
it should be made cheaper. Please remove static initializer and any work in 
constructor that you do that. 

How often does this happen are there any particular steps to reproduce? I can't
reproduce.

I am attaching a patch which removes the static initialization, and the
resulting module. I would prefer if this was tested by someone who can reproduce
this and can verify that it helps.

BTW, looking at the stack trace, it seems overly complicated. The .settings file
format is well known, so couldn't just SharedClassObject directly parse the
settings file and get the result? Does it need to go through the
Lookup/FolderLookup/XMLDataObject layers?

Created attachment 26406 [details]
Patch in httpserver module

Created attachment 26407 [details]
Module with the patch

Fixed. 

Checking in HttpServerURLMapper.java;
/cvs/httpserver/src/org/netbeans/modules/httpserver/HttpServerURLMapper.java,v 
<--  HttpServerURLMapper.java
new revision: 1.12; previous revision: 1.11
done

dbalek,could you please verify the fix? Thanks.

Verified.