Good move!
This points out an extremely important item, and one that is often overlooked: the coaxial and optical representations of S/PDIF/ADAT aren't _quite_ the same items, even though the exact same _bits_ are being sent. The coaxial form is sent over 75ohm coax with proper line drivers, and it has *enormous* bandwidth (several hundred MHz if done with care, and certainly at least 50MHz even if done by a complete wanker): doing 2 channels of 24/44.1 is an absolute walk in the park for both the send and receive hardware, since you only need 1.15MBit/sec to do it.
Optical is another story, on the other hand. The TOSlink hardware (the transmitter and receiver devices themselves that launch the optical signal up the fiber and receive it) was designed for 2 channels of 16/44.1, and there's very little bandwidth margin above that requirement. Everyone always thinks of fiber has being higher bandwidth, and it certainly *can* be: but those particular consumer items were designed for absolute minimum cost, not any real performance target.
Here are the datasheets for the Toshiba TOSlink stuff (they originated the hardware, and are by far the most commonly used):
http://www.toshiba.com/taec/components/Datasheet/TOTX173.pdf and
http://www.toshiba.com/taec/components/Datasheet/TORX173.pdf
Max data rate is spec'd as 6MBit/sec. Let's see: 8 channels of 24/48 data is 9.22MBit/sec: so you are running the hardware at 150% of its design max *right off the bat*...
Yow. No wonder cheap optical patch systems have problems with ADAT. In order to make that work at all, IMNSHO you really need to receive the data, reclock it, buffer it, and _then_ send it out with corrected timing and framing. In my opinion you cannot just connect a transmitter behind a receiver, shove the raw signal back out again without reclocking, and just hope for the best... Running that far above the bandwidth spec has to smear the edges of the pulses all to hell and gone in the time domain, and doing it twice back to back would leave you with _mud_. I suspect that that is why the Midiman design (and others) have problems with Type II ADAT: after you go through 2 send/recieve pairs, the lower bits probably fall so far out of frame sync that they go the way of the dodo. It'd be interesting to try looping through one of those 2 or 3 times, and seeing how many bits might still be seen wiggling on the other side of the final receiver! The cumulative timing errors would be interesting...
Anyway, what the Z-Sys box does is all that resyncing and buffering, and that's why it is more expensive.
So if you want to do _really_ cheap S/PDIF switching, then doing it in the 75ohm, coaxial domain certainly makes much more sense: the extra bandwidth buys you a ton in terms of data integrity, and therefore reduces the need for magic to be done to preserve it!