I believe Ken estimated it back on 20 May 2008. You might search Rootsweb DNA L Archives for other estimates. The 3,500 figure you cite for R-21 seems to fit in nicely.
http://archiver.rootsweb.ancestry.com/th/read/genealogy-dna/2008-05/121131164420 May 2008
"it now automatically now gives time back to the MRCA of both clades as well as the MRCAs for the separate clades. I tested it with S28 against M222 as well as M222 against S21 I had to eliminate 481 and 413a from the sums because it appears a multi-step single mutation occured there in M222"
G(M222) = 1740 years
G(M222/S28) = 3780 years
G(M222/S21) = 3570 years