I noticed an interesting relation between U3 and U6. if you compare the concurrent data sets avoidable from 1994 to Jan 2009, you will find that U6 = U3* 1.77. U3 is related to U6 by a factor of 1.77 There may be some underlying cause for this relation that is associated with the manner in which these two data sets are generated.
Regardless, this relation should allow us to backcast the U6 data which starts in 1994 to 1948 when U3 data begins.
Here are 2 charts showing what i call Adjusted U3 (U3 x 1.77) vs U6. Visually they are very close as well as statistically. The Standard deviation for the the data set of U6/U3 is 0.06
U3 Backcast using the 1.77 factor