The other situation in which it's a "huge, huge problem" is when there is communication they'd like to access that is encrypted.
What we don't know, among many other things, is: 1) How often do they come up against encrypted comms? 2) Can "traditional" work yield results that would make the encrypted comms less vital to the investigation? 3) What's the scale of the crime they're investigating when encrypted comms are encountered? 4) Are they investigating after a crime has already been committed? If so, then the damage is already done. (Yes, reconstruction of the lead-up may be useful)
Breaking encryption is not going to be a silver bullet, but trying to sell it as such is their "huge, huge problem".
Storing and sifting mountains of data and metadata from everybody, and mining it and mapping it, just waiting for somebody to do something wrong is just too tempting to resist.
That said tho, I tend to believe that most encryption is crackable due to mistakes made during implementation, or mistakes made by the user. In a cryptanalysis class I took my mind was blown by how easy it an be to decode cipher text using simple statistical analysis, especially if some of the plaintext is known.
The problem (for him) is that encryption is a tremendous help to everyone else, and we don't (yet?) have a way to protect our data from criminals and from overzealous or corrupt cops, without also deterring legitimate investigations.