Request for review: 6806649

[Christopher Hegarty - Sun Microsystems Ireland]

6806649: synchronization bottleneck when constructing Thread subclasses

Webrev:
   http://cr.openjdk.java.net/~chegar/6806649/webrev.00/webrev/

The global synchronization on Thread.subclassAudits at Thread 
construction time has been discussed as a bottleneck:

http://cs.oswego.edu/pipermail/concurrency-interest/2009-February/005839.html

The implementation approach for this cache of subclass audits matches 
the once-similar implementations in ObjectInputStream and 
ObjectOutputStream, but since the 5056445 fix in JDKs 5.0u7 and 6, those 
classes were changed to use a ConcurrentHashMap (with weakly referenced 
Class keys) instead-- Thread should probably be changed similarly.

Thanks,
-Chris.