Work array for parallel loop

[lmiq]

Yes. That or allocating a vector of work vectors outside the threaded loop (if you want to preserve that allocation).