Which isn't to say that this is bad work, nor even that the authors are over-hyping it (in fact, the paper mentions Spectre... very early in the introduction only.) But this article makes it easy to over-estimate the impact.
As always, crypto code (in particular) must be constant-time (constant-everything) to avoid key recovery via such attacks. But again, that isn't new - cache-based attacks have broken lots of non-constant-time crypto code.