Can we override the scheduling used in Cilk if for our application purposes it would hurt performance?

How is the work queue implemented on a per thread basis? If each queue is stored in each core's context, would there be a significant performance penalty for stealing from another worker's queue, potentially eating up clock cycles to copy over work from a different part of the CPU?

@jtguibas I would assume you'd instead use a different form of parallelism if you know from the outset that you'll have lots of small tasks or another workflow that would be hurt by this.