Silverpig is right. If you draw a circle on a sphere, then assuming that the center of the circle lies on the surface of the sphere (which is pretty fair, otherwise it wouldn't be called drawing it
on the sphere...) then the radius of the circle suddenly becomes much larger.
So say you do this, you draw a circle on a sphere. Then the center of that circle would be on the surface as well, kind of like the white(ish) dot on this picture:
So then taking out the wireframe of the sphere, and just leaving the radius, circumference, and a diameter, you get something that looks like this:
In other words, a cone.
Now, since the radius of our 3D circle (which now happens to be the slantHeight) is longer than the Radius as depicted in the cone, the ratio between the radius and the circumference is decreased, i.e. it is no longer 3.14159265358979323... but something appreciably smaller. In fact, by modulating the position of the circle in the sphere pi can be made smaller than 1.
Obviously I have made a few shortcuts for simplicity; in reality, taking the straight-length from the center to the circumference, as I did, in fact lessens the effect; running the radius along the outside of the circle, since it is a longer distance, would have an even greater impact on pi.
However, this obviously only works in a reality where a circle is our plane equivalent; i.e. it won't work in this universe, or at least it won't under our current conditions. Which is why pi is known as a
constant.