Eyeballing the figure at http://en.wikipedia.org/wiki/Fluctuations_in_the_length_of_day#Observations http://en.wikipedia.org/wiki/Fluctuations_in_the_length_of_

Eyeballing the figure at http://en.wikipedia.org/wiki/Fluctuations_in_the_length_of_day#Observations http://en.wikipedia.org/wiki/Fluctuations_in_the_length_of_d... it looks like you could expect stellar time to be off proper time by not more than around 0.1 seconds in 100 days, since the width of the band of fluctuations is around 1 ms. (I'd say 0.01 seconds since in a random walk the error grows with the square root of time, not linearly.) Maybe this fluctuation graph has subtracted out seasonal variation?