Someone else pointed out the most likely accuracy issue - Google may (probably does) use a dynamic balancing system to add and remove servers/nodes according to load.
The checks I ran were done over just 4 days. The resulting numbers may have no relevancy at all over other time periods with different load amounts.
I don't know much about the German tank method but it seems interesting that given 62 unique addresses with a maximum of 103 it would give a total number of 104. This doesn't seem to make sense to me. Can someone explain why this works this way?
It takes into account factors unique to tanks, so you'd have to define your own distribution for Goolge's gadget servers to make a correct estimation.