The only logical way to run the network is scheduled departures (since you cannot actually schedule the arrivals), and the canonical X-Mail is digital only. Canon also has the pilots swapping out, as well as the ships. And even with daily departures, it's still better for the network overall to depart on schedule, because few X-boat stops don't have multiple destinations served; what is there goes on schedule. What isn't waits for the next one going.
It is not the "only" logical way, it is
one logical way which you prefer for your stated reasons. I, OTOH, see it -- compared to a "best effort" unscheduled model -- as inefficient from a systems management viewpoint, not to mention being counter-canonical from a practical one.
By way of example, I used to have satellite internet service, and (distance to geosync orbit and back notwithstanding) the ping times were atrocious due to the many-millisecond scheduling delay built into the transmitters. My ping times were routinely nearly two seconds long, only about half of which was actual travel time up to and down from the "bird" in orbit. The rest was from packets sitting in a buffer in a transmitter, waiting to go out on a schedule, and seeing if any more packets might arrive before then, that they could all go together, or otherwise leaving without them and letting them show up late and have to loiter in the buffer until the next scheduled "squirt" time rolled around.
Extending beyond this issue to the matter of unscheduled arrivals, my point is that you cannot schedule the departures
without slowing down the throughput if a packet is late in arriving -- which will either happen about half the time due to Jump uncertainties, or else you make each Jump leg average 7.5+ days long instead of 7, since with the scheduling you are using, a late-arriving Xboat sends its throughmail on down the line on the
next scheduled departure tomorrow (and the Xboat that left on its daily departure schedule a few hours ago might well be running empty except for any local data uploaded from the mainworld and headed on down the line), but this extra half a day on average adds a transmission delay to the whole system -- especially over long, multilink routes, and the packets are not "always moving at top speed" as per canon.
And that is what this hinges on: for data to move down the Xboat line as fast as possible it needs to spend as little time as possible sitting in the data banks of an Xboat that has not Jumped yet. Otherwise, that data is as idle as the Xboat carrying it, waiting on an arbitrarily-chosen point in time to roll around before it starts moving again.
The operational model of the original Pony Express is cited in canon only inasmuch as waiting-horses-and-riders who galloped off the moment the courier bag had been handed off is put forward as being exactly analogous to the way Xboats operate. The details of how horses and riders were accommodated at the stations and other practical operational parameters of historical significance make up a separate and non-germane issue from how the service and its couriers were operationally employed. It is -- if you will pardon the metaphor -- a dead horse that we do not have to beat here. We can take it as read that XTs are kept busy processing Xboats and pilots to get them turned around for the next departure, and that this process is time-consuming and results in a lot of logistical juggling involving numerous personnel and vessels (including all the Type S hulls being dispatched for off-route deliveries and pickups), even for a quiet, two-link-only hub node, or a one-link-only leaf node. There is obviously a good reason that Xboat hub nodes rarely have more that three links/routes converging, and almost never have more than four -- the local operational logistics would resemble a punishment detail. Your basic numbers for assets-in-system seem about right to me, as per Orr above.