Consider a class C of possible universes where, at t0, exactly 100 independent random processes are activated, and nothing else random happens. Suppose there are no qualitative differences between universes in C other than those due to the differences in the outcomes of the processes. Suppose, further, that each process can result in either a "heads" or a "tails", with equal probability 1/2.
There are, thus, exactly 2100 different types of possible universes in C (where we say that universes are of different type provided that they're not exactly alike).
Suppose we live in a multiverse that contains exactly n universes from C. If n<2100, then there are some possible types of universes not represented in our multiverse—there are some combinations of heads and tails in a C-type universe that don't occur in our multiverse. If n>2100, then there are some types of C-universes that are multiply represented in our multiverse.
So if every type of universe is represented exactly once, there are exactly 2100 C-universes in the multiverse, and each has a different heads-tails profile. But how likely is it that each would have a different heads-tails profile? Assuming that what happens in different universes is stochastically independent, this is just a version of the birthday problem. If n=2100, then the probability that each of the n universes has a different one of the n profiles is n!/nn, which according to Stirling's formula is something like exp(−2100). That's a very, very tiny number. And since classes like C can be found for any number of random processes, not just 100, it follows that the probability that every type of universe is exemplified exactly once in the multiverse is smaller than every positive real number, i.e., it's zero or infinitesimal.
So if we live in a multiverse, almost surely either some possible types of universes aren't instantiated or some are multiply instantiated or both.