i can't remember the specific track widths (I'll check) but I didn't use anything special above the standard traces used for the key matrix. In either case the resistors (I have one per LED, all arranged in parallel in LED/resistor pairs) would give you greater resistance than the traces. I also added a rheostat in series with the LED clump to control the power going in to be able to adjust the current in case the LEDs get too loud. Full size, 106 keys, bright white LEDs all around.
In terms of how best to run the traces - I kept all traces on one side of the plate as much as I could (switch matrix, 5V traces for LEDs), to be able to have a mostly uninterruped ground plate on the other - I then just wired all the LEDs with direct traces to the 5v, then just dump the other end into the ground plane. Saves running the LED traces back to the controller.
NB: I used a Teensy++ using one 5V output, and connected (2) GND connections to the ground plane. Works swimmingly for me. I designed for 100Ohm resistors + 200Ohm rheostat, I kept being sent 1kOhm resistor so used those in the prototype to see how they worked. Full brightness is fine but I'd like to get them a little brighter. I haven't tried the 100Ohm yet but I get the feeling max will be waay to loud. Hence the rheostat to tune - LEDs are apparently finicky and vary even within batch.
EDIT: re-reading your post I see you're using RGB - this obv. complicates the traces above what I did which is single colour static. Full RGB for this many sounds like a routing nightmare. If it were amateur me I'd put as much keyboard matrix on one side of the board, then all the LED RGB on the other, then weave back to ground on the KB matrix face. I don't think you could do a ground plane unless you go three layer which is above and beyond what I know about dealing with. Keeping the split distinct, neat, and tidy.
EDIT2: LEDs are rated for max brightness - you can run them at much lower current (at least one-colour) for comfortable keyboarding. Full is way too much anyway. I was figuring, iirc, at 2mA per LED, though I'd need to double check my notes to confirm.