Take the kth car. It is at the start of a cluster if it is the slowest of the first k cars. The kth car is therefore at the start of a cluster with probability 1/k. The expected number of clusters is the sum over all cars of the probability that that car is in the front of a cluster.
Hurray for the linearity of expected value!
If it's worth saying, but not worth its own post (even in Discussion), then it goes here.