Visual Studio 2012: Slow deallocation when new called within openmp loop

#include <iostream>
#include <vector>

#include <time.h>

typedef std::vector<double> DoubleVector;

int main()
{
  int nhist(70000);
  int nbins(2000);

  std::vector<DoubleVector*> dataY(nhist);
  std::vector<DoubleVector*> dataE(nhist);
  std::vector<DoubleVector*> dataX(nhist);

#pragma omp parallel for
  for(int i = 0; i < nhist; ++i)
  {
    dataY[i] = new DoubleVector(nbins);
    dataE[i] = new DoubleVector(nbins);
    dataX[i] = new DoubleVector(nbins);
  }
  clock_t start = clock();
  for(int i = 0; i < nhist; ++i)
  {
    delete dataY[i];
    delete dataE[i];
    delete dataX[i];
  }
  clock_t now = clock();
  const float retval = float(now - start)/CLOCKS_PER_SEC;
  std::cout << "Time for deallocation=" << retval << " seconds." << std::endl;
  return 0;
}

The attached code demonstrates an issue that we have observed within our application that has been compiled using Visual Studio 2012 with Update 3.

Allocating a large amount of memory while inside an OpenMP loop causes the deallocation time to increase quite severely. Running the above example code, in release mode, takes ~1s to deallocate all of the vectors on my system. If I comment out the '#pragma omp parallel for' line then the deallocation time drops to ~0.3s.

Furthermore, if I increase the value of the nbins parameter to ~2500 then the deallocation time is ~340s!

Can anyone offer an explanation for this significant difference?