It was my impression that Wirth's law was mostly intended to be tongue-in-cheek, and refer to how programs with user interfaces are getting bloated (which may be true depending on your point of view).

In terms of software that actually needs speed (numerical simulations, science and tech software, games, etc.) the reverse has always been true. New algorithms are usually faster than old ones. Case in point is the trusty old BLAS library which is the workhorse of scientific computing. Modern BLAS implementations are extremely super-optimized, far more optimized than older implementations (for current computing hardware, of course).