Use this Python code to compare SciPy speed to our benchmark. Yes, it's varying wildly.
from datetime import datetime
import numpy as np
from scipy.ndimage.filters import gaussian_filter
# Warp up
a = np.zeros((134, 156, 130), dtype='float')
g = gaussian_filter(a, 1.0, truncate=2)
# Time
start = datetime.now()
a = np.zeros((134, 156, 130), dtype='float')
g = gaussian_filter(a, 1.0, truncate=2)
print((datetime.now() - start).microseconds / 1000, 'ms')