The propagation delay of a signal through multiple load devices coupled sequentially along a conductor is improved by separating a subset of the load devices that is more distant from the signal source, and coupling the more distant subset to the signal through a fly-over conductor that bypasses the subset that is nearer to the signal source. The technique is applicable to subsets of bit cells in a random access memory (SRAM) coupled to a given word line, or to word line decoder gates coupled sequentially to a strobe signal, as well as other circuits wherein load devices selectable as a group can be divided into subsets by proximity to the signal source. In an SRAM layout with multiple levels, different metal deposition layers carry the conductor legs between the load devices versus the fly-over conductor bypassing the nearer subset.