The story (blog article) refers to this short essay:

http://download.cell.com/images/edimages/CurrentBiology/home...

which mentions "The long-lasting effects of TDCS, which can persist for up to 2 months" with a reference to this paper:

http://neuro.cjb.net/content/29/22/7271.full.pdf

This last paper is quite interesting, as arguments can (will...) be based off of it posing that tDCS does provide a long-term benefit. Unfortunately, the gain (in response time and accuracy) over a sham tDCS (e.g., placebo) is quite modest -- 15% with an error bar of 10% for response time, and 5% with an error bar of 5% for accuracy... Check the Figure 4 on page 5.