Currently jlib maintains a list of active threads which is used for displaying in the stack trace and also in the process monitor for mapping busy threads.

Maintaining this list adds an overhead

• critical sections at the start and end

The critical sections don't add very much

• maintaining the array. 

If there are large numbers of threads then using an array is inefficient because removing them  becomes expensive (3,000 threads ~2us 30,000 threads ~15us).

See branch ghalliday/activeThreadLinkedList for an implementation.  It seems to slow down small  threads a little (but hard to tell because of variations in timings).