There is a way of looking at this phenomenon in terms of an optimal path problem, Fermat's Principal of Least Time.

Suppose you have the following situation

    x1  |  n2
        |
        |
    n1  |  x2

We have two different materials, with index of refraction n1 and n2. Now suppose that light has traveled between points x1 and x2, we can ask the question, what path did it take? (Note here, we are asking about classical rays of light, so we're thinking in terms of geometrical optics, not quantum :) )

When light travels through the left hand side, it's traveling at a velocity, c/n1, but in the right, it will travel at c/n2.

It turns out that we can answer the question of which path the light takes by asking "Which path takes the shortest amount of time?". From that assumption you can actually re-derive Snell's Law.

An example when n2 is very large (and thus moves slower on the right hand side):

    x1  |  n2
      \ |
       \|
    n1  |--x2

the light will travel more distance on the left hand side because it can move faster there, and thus get close as possible to the destination. It will then make the last bit of distance in the right hand side.

Now Snell's law says

  n1 sin θ_input = n2 sin θ_output

Since n2 is very large, the only way to make this work is to make sin θ_output very small. So we see in this case at least, both approaches agree as to what will happen.

So it turns out this principal works even in the case of gravitational spacetime distortion, and can be applied there as well.