According to the trials from White and Zainasheff's Yeast the yield factor starts falling around 1 vial / 2L (50m/ml), but is still fairly high at 1 vial per 8L. In a 1.036 simple starter, yeast will double 1.1x for 1 vial in 2L, 1.8x for 1 vial per 4L, and 3x in 8L. You can see how doubling increases at a lower rate than volume.
Yeast growth is sensitive to pitching rate. High pitching rates will result in propagation-type growth, low pitching rates will result in beer-like growth. The growth-vs-starter-size curve gets flatter and flatter the larger your starter gets, and approaches a limit of 600b cells, regardless of wort volume. We can use this info to our advantage when planning yeast starters.
For instance:
A) 1 vial = 100b cells > into > 8L = 400b cells (3.0 doublings, plus 100b cells we started with).
For a 2 step starter:
B) 100b cells > 2L = 210b (1.1 doublings) > 2L (0.5 doublings) = 315b total
C) 100b cells > 2L = 210b (1.1 doublings) > 4L (1.1 doublings) = 441b total
For 3 steps:
D) 100b cells > 2L = 210b (1.1 doublings) > 2L (0.5 doublings) = 315b > 2L (~0.25 doublings) = 394b total
Even though C and D had the same total amount of wort (6L) the growth in D was lower because of the increasing pitching rate at each step limiting the growth. Even though C used 2L less wort, it grew 41b more cells. C had 25% less wort, but grew about 10% more yeast. D used 25% less wort than A but grew almost as much yeast.
Now, for simplicity's sake it may be more practical to just make one 8L starter, but if your glassware isn't large enough, or if you're using DME to make your wort ($$$) then perhaps doing multiple steps makes more sense.
EDIT: As Tom pointed out, the term "doubling" that White and Zainasheff use is actually a multiplication factor, not a true "doubling" figure.