The reason it's OK to extrapolate in this way would appear to be in this fact (from TFA):

"Unlike other prominent planet-detection techniques, which measure the shadows of planets passing in front of their stars (transit) or measure the wobble of a star due to the gravitational tug of a planet (radial velocity and astrometry), the gravitational-lensing technique is unbiased in the selection of the host star."

To know more, we have to read the story in Nature, which is here:

http://www.nature.com/nature/journal/v481/n7380/full/nature1...

Look at the error bars on their percentages. Super-earths (planets of 5-10x earth mass) are present on "62 +35 -37"% of systems. So, your intuition that the extrapolation was shaky is correct.