To save your computer the effort of cruching numbers (and for whomever wanted more math)...

The data matches a normal (Gaussian) distribution exactly. So to make this chart:

BB/100 percent
-5 4.74
-4 5.96
-3 7.12
-2 8.10
-1 8.73
0 8.97
1 8.73
2 8.10
3 7.12
4 5.96
5 4.74
use:
percent = 100*i/sqrt(2*pi*s) * exp(-b^2/(2*s))

where:
i = BB/100 increment (i = 1 for 10k, 0.2 for 50k, 0.1 for 100k)
s = 200k/sample set size (s = 20 for 10k, 4 for 50k, 2 for 100k)
b = BB/100 (first column)

s = sigma^2 in the guassian formula. 2*sigma represents 95% of the population, so set:
v= acceptable BB/100 variation (+-1 for Pyroxene)

then:
Desired sample set size = 200k/v^2

by Pyroxene's method and thinking:
v = 1 => Desired sample set = 200k

If +-2 BB is acceptable then 200k/4 = 50k is a satisfactory sample set.

A sample set of 10k implies a possible variation up to +- 8BB (by Pyroxene's setup)