Lambda switches bear on any plans for heavy duty networking. The general scheme is to switch whole (light) frequency bands without buffering. I cannot find any detailed quantitative information about them. Here is an early mechanical Lucent switch in the form of a steerable mirror for light beams presumably from fiber via a diffraction grating to mirror to another fiber. It looks like it could switch in milliseconds. Deployed in Japan in 2000 and across the Atlantic in 2001.

This note suggests that they can switch the color of a stream as they recombine streams from different incoming fibers to form an outgoing signal for another fiber. I know no physics that can change color except for integer multiple of frequency. There is no hint on whether the switching is done on a one second or one nanosecond time scale but this paper proposes a technology that somewhat dynamically routes whole colors and presumes μs scale color switching. Such a technology would be strategic to bandwidth management on the 100 millisecond timescale but that would presume some centralized control. These ideas presume out-of-band switching signals since the digital content of the switched stream is not available at the switches. The crude scheme is to schedule ahead periodic durations of less than a millisecond when the stream at some particular color (lambda) will be switched to some particular fiber. This is reminiscent of the ideas in the successful but dated ‘time-slot-permutation’ of SONET switching.

This 2001 paper talks about ‘light path setup’ and assumes but does not explain ‘wavelength converters’.
This 2012 note claims that there are no commercial pure optical color converters yet.