I think the longest SPI chains you can find in LED displays. We are running 10x16bit shift registers at about 2MHz in one of them. Nevertheless, there is no return, as the display does not respond in any way.
In an another design, we chain 2xin + 2xout 8-bit registers, but they are placed remotely from the controller board (cca 1.5m) - that required a flat cable with each other line grounded (or better an FTP), and some filtering and clock recovery/buffering (at approx. 1MHz).
Usually, as the data is fed only to one device, only the clock needs (re)buffering after a couple of stages (and sometimes also the latch/synchro line, but usually it does not mather that it is slow).
The return is sometimes a trouble too, purely because of physical length - one tends to stretch the line liearly...

Jan Waclawek