Okay, it's been a pretty bad year. It turns out the Maestro application wizard won't do code that would actually compile with MPLab IDE, and googling on the problem I found a similar issue posted on Microchip forum, post dated 2005, with no help or replies. So no LED testing yet. There is a wide variety of LEDs with different brightness available though, so it wont' be a problem.
Anyway, I've been doing some considerable rethinking on the system. Allegro discontinued their 32-bit LED drivers, but offer newer 16-bit drivers with some very nice features, and Maxim offers really nice serial drivers as well.
The timer I made for the ASCII display turned out to be a vain effort; upon closer reading of the datasheet, it plainly stated out; "Designers not wishing to use Read function may tie pins !CE and !WR together". So that's a big bonus. Also, I could emit both pins A3 and D7 (they are only needed for user defined characters), further helping with the wiring.
Going with sink driver only meant reversing the LEDs on the display boards and changing the displays to Common Anode style (though, matching common anode 3x7 and 4x7-segments are much more difficult to find than common cathode for some reason).
Anyway, here's the revised display board schematic:
Reconfigurable pins have been reduced to minimum. The original setup was wholly unnecessary and createed an impenetrable mesh of wired on the board. This is simple, nice and offers good forward compatibility with possible future revisions.
Here's how it worked out as PCB:
The thick traces are common anodes for the LED arcs. Going by the number of LEDs and their highest allowed overdrive setting, the highest current they will see is 4 A, but only at about 2% duty cycle (if even that) and only when the leds are driven at full brightness at maximum display value. Any case, LEDs will burn out LONG before the traces. However, running 103 leds per board can cause a bit of heating, I tried to arrange the wires so it'd be as even as possible.
The PCB mask image shows the LED arc traces best, so here's a shot:
I personally like how the common anode traces turned out, the one running between pads is a bit thin but I can hardly imagine it causing trouble. The only thing I do worry is the two anodes running directly on top of each other at the outer edge.
I've tried to minimize the number of vias on the board, as there's already more than enough holes on it.
There's a 3D package for Eagle, but since it uses POVRay which is a write-only format and one of the worst examples of complacent circle jerk software available, I really have little need or interest to bother with it.