I explicitly emphasized the following:
in step b the car picks up 4 first and then 5, <-- IMPORTANT!
It is true that at a * state, 5 is falling behind. But after step a and when we get into step b where the car is coming back to meet 4 and pick him up, it must be the case that, 5 has already passed 4.
And this is exactly the reason why from one "* to * iteration" to the next, the distance that 5 is following behind is reducing at a constant rate r, where r must greater or equal than 0.5357, and it cannot possibly reduce any faster than that. This magic number is computed for the case that, when the car comes back to meet 4 and pick him up, 5 is right catching 4 at the same point.
Again
(a) as long as in step b, when the car picks up 4, 5 has already passed 4, the speed of the 2,3 and 4 will be the same as the 4 people case (in fact, it's easy to see 5 can be ignored).
(b) while (a) satisfied, the distance that 5 is falling behind is reducing at a constant rate r, where r >= 0.5357. That means, while 2,3 and 4 are advancing at a constant speed, if in one iteration 5 is falling behind d, then the next iteration he's only falling behind d * r. As long as r is less than 1, it converges and 5 is catch up with 2,3 and 4 the end of the day.