In order for the arithmetic mean to be integer, the difference between $a$ and $b$ must be even, and in order for the geometric mean to be integer, the difference between $a$ and $b$ must be some multiple of a difference of squares.
This means that for a working pair $a<b$, the difference can't be less than $4\sqrt a$, because the difference between adjacent same-parity squares will be at least this much, and working at multiples (>2) of the difference of squares will increase it further.
In order to prove the proposed minimum difference of $30$ in Ross' answer, therefore, it's enough to check pairs with the smaller value up to $56$, which I have done.
Of course, the limit there is hindsight. I actually checked numbers up to 40000 and reported back the 48 cases where the resulting calculations were co-prime. In each case this "base pair" was created by taking two small squares, multiplying both by two if necessary to get the difference even, then multiplying both by the average of the two. $(5,45)$ is generated by $(1,9)$; $(10,40)$ comes from $(1,4)$. The original pair found in the question is a doubled version of $(5,45)$, of course.