explicit threads_stack_limit (for recent Poly/ML SVN versions), which leads to soft interrupt instead of exhaustion of virtual memory, which is particularly relevant for the bigger address space of x86_64;