Simple answer:  we don't know because AM's hash rate does not conform to any common distributions.  Also I am quite comfortable admitting that it is a difficult problem, and other posters on this board could probably make better calculations than me.  IMHO Admitting your own limitations is perfectly OK...

But I said 4 posters weren't thinking critically enough, but I was not counting you because I believe your calculation is reasonable.  You were the only one to recognize the concept that a GIVEN 6 hour period is different than ANY one 6 hour period. 

You assume a distribution function that is not accurate, but it is reasonable.  Realistically, given that we know that AM's hash rate is varying (sometimes it is higher and sometimes lower as friedcat as mentioned several times that blades occasionally go down), your percentage is probably too low.  But really from any given hash rate estimation (ie 10 blocks found in 6 hours), all you can do is construct confidence intervals about the hash rate.  For example, we can say things like, AM's hash rate is likely between 30 and 60 THash/s for a X time period.  As you get more data you can improve the accuracy of the hash rate, but the problem with collecting more data is that you now you have an issue with nonstationary data. 

We will never know what AM's hash rate was exactly over a given time period, but we can construct reasonable ranges.  Just all the pseudo-intellectualism and misleading of other posters kind of infuriates me, so I apologize if I come off too strong.  Its just that the smugness / condescending posts bother me.