Hi Kevin,
The reference signal is necessary to ensure that the router switches occurs at a predictable time. Essentially, the router operates "blindly", it has no idea what the individual signals are, or how they are timed. It simply performs all switching at a fixed time, relative to the reference signal. There is a configuration setting related to it, which tells the router to use SD, HD, or a custom switching point.
It is assumed that all incoming video signals are also timed to the same reference. If they are not, the router will still switch, but it will happen at an arbitrary point in the video signal. For example it could happen in the middle of active picture. Downstream equipment may tolerate such a switch, or it might flag a "loss of input" condition. Either way, the picture would be incorrect for at least one frame.
The SDI video specification defines a specific point during the vertical blanking as the "switching line". It is acceptable for this line to be longer or shorter (because individual video sources are never going to be perfectly aligned). No other data is normally encoded on this line, since it may be lost during switching. If you are interested in the details of this, have a look at SMPTE RP 168.
I hope this helps!
-Ralph
#NK