The only reason to limit ourselves to primes is because its trivial to produce a meaningful composite, and very difficult to produce a meaningful prime. Take this message; if I was to associate it with a number (say by joining its characters as 7-bit ASCII), my money is that it's composite. (Not sure what the odds are, but I'd take 100:1 odds).
EDIT: period is `01110`, which is even, so I win.
Can we estimate how many gigabyte sized primes exist? I think one can do that using the prime number theorem. Let’s call that N. What is the probability one of those N messages is meaningful? I think it is a lot higher than you think it is. I think N is a very big number, even though it is very small compared to the number of composite gigabyte-sized numbers, it is still unimaginably large in absolute terms.
I mean, it's not that many if your expecting a message from God (I say half-jokingly).
Much more interesting is to discover a new communication, a message encoded in a very large number that is distinguished by being prime. (What message would you send? It would be interesting if it ultimately described some sort of recursive, reactive automata, the source code for a kind of learning, communicating intelligence. A gift, then. An awesome, terrible gift.)
Any finite message (containing a finite number of symbols drawn from a finite alphabet) can be represented as a positive integer.
Any positive integer n can be encoded in a prime, even a composite integer.
Use the following encoding system: write 1 n times, followed by a zero, followed by an arbitrary bit string.
For example, for 4: 11110xxxxxxxxxxxxx...
Using that encoding system, there will be an infinite number of primes that encode 4 - all the primes which start with 11110 in binary.
Thus, the set of all primes contains an infinite number of copies of every possible finite message.