[yarg]

I find their assertion somewhat disingenuous, in security things are often done (and known to be and deliberately done) in rather inefficient manners - for example algorithms that run in fixed time despite obvious case specific optimisations.

We'd probably be far better off coming up with mechanisms to generate strongly probabilistically unique psudeo-random domains from which we could (on a per cryptographic context basis) generate strong pseudo-primes.

This could most likely be done in a reasonably efficient manner, without resorting to the government backdoor suggested.