I thought something like that was the case, but couldn't be bothered to go back and check.
I've seen that "conveniently clustered at one end" before in amateur constructions. A close relative is to cluster all the decoupling capacitors where the PSU enters the board; might be worth checking that!
Point out to the client that a single output driving
100ohms through 5V requires a current of up to 50mA until the reflections have died out. With source termination that is reduced to 25mA. That's within the spec for 1 pin, but not if all eight outputs change simultaneously. However, if the source's transition time is slow compared to the transmission line's length, that will be less of a concern.If the obvious simple cable connection doesn't work /reliably/, then the possibilities are limited and "sub-optimum"[1].
Educate the client in the costs of noticing/determining/fixing unreliable behaviour in the field. That may involve formal documentation to the effect that you cannot be held responsible. Sometimes you have to decline to do work if it will be too much trouble in the future.
Consider having a "very fat" connectors which do a least inappropriate combination of: - rearrange the grounds more appropriately. - if there is an obvious clock, preferentially improve that line's impedance/crosstalk by surrounding it with gnds. - insert source termination resistors in the box - for non-clock signals, add capacitors after those resistors to slug any rapid transients - have receiver termination resistors to prevent over-voltage slowly knackering the receiver's input protection diodes (check the rx spec)
If the client doesn't believe you and has a short attention span, point him to the relevant "Bogotin's rules of thumb".
[1] code word for lousy