The appearance of the plot will be affected by the window types and durations used to generate it as much as the smoothing applied, so you need to make sure the windowing is the same before looking at smoothing comparisons.
After getting the windowing the same the difficulty is that "1/3 octave" really only specifies the frequency span encompassed by the smoothing applied but does not define the smoothing kernel. A simple moving average that has a 1/3 octave span will appear much less smoothed than a Gaussian kernel used over the same span (in fact a Gaussian is well approximated by multiple passes of a moving average). REW's smoothing is approximately Gaussian.
ARTA uses a 6th order Butterworth filter on data that has been converted to logarithmic spacing, from the manual:
Note: Preceding examples show 1/3-octave smoothed curves. The smoothing of frequency response curve is done in different way than smoothing of spectrum magnitude. The ARTA uses approach to first interpolate and average frequency response on log-frequency axis, then makes smoothing by convolving that response with response of bandpass 6-pole Butterworth filter.
With the same windowing you should find the two approaches look pretty similar.