I think that MBLUR could use any number of samples, but it is limited to those options in blender.
Is there a reason why blender has those values for OSA?
For the jittering in OSA (and MBLUR [it uses jittering too]), maybe the samples are taken in certain arrangements...
e.g.
5 - it could be a square of 4 samples plus one sample in the middle.
8 - maybe a 3x3 square with the sample in the middle missing.
11 - I don't know what that would be
16 - a 4x4 square...

I though it would just have 4, 9, 16, and 25 as the options...

I think the trick of jittering is that the distance from 1 point to the surrounding points is the same.
These are precalculated and stored in blender for the numbers 5, 8, 11 & 16.
These numbers are 8 years old, for modern machines 8, 11, 16, 25 might be better suited.

The surrounding points....?

What about "11"? What are the 11 points? A 3x3 grid plus two on the sides?

Why did they use "8" rather than "9"?

BTW, during jittering, does it actually rerender the image a couple times or does it just combine a few copies of the image?

For me, the time taken to do the OSA seems to be about proportional to a square root of the number. e.g. OSA 16 seems to take about 4 times the amount without OSA or about twice the amount of OSA 5.