RGB SMT LED Cube, resistors, drivers, and shift registers.

I did a lot of soldering yesterday. I found some unused xmas lights that had wires in groups of 3, about the right lenght for my project. I cut, stripped, tinned, and soldered them onto the resistors that I had previously soldered onto the column wires on the cube. I ended up having to untwist the wires to move them through the case, and it ended up pretty ugly. I will probably end up replacing those wires, but they will work for now.
Before green wires:


After green wires:

The horror!


Soldering away...

Which brings us to here:

Now that I have the shift registers wired up to the ribbon cable (clock, latch, data, ground), and wired to the LEDs, I need to figure out how to source current to the common planes.

In my first cube, I used 20 pins, 4 pins sourced 5v (through 220 ohm resistor). I upgraded that cube to using 2 shift registers to control the sink, and sourced 5v via 4 pins through 220 ohms.
I recently upgraded that cube to 16 100 ohm resistors on the column(sink) pins, and removed the 4 220 ohm resistors on plane(source) pins.
In my RGB shiftpwm tests, I sourced 5v to the common pins via the 5v pin on the uno (Originally I had it from a 5v power supply, but later I moved it to the uno for portability).

It seems to me, each of those are drawing too much current through the arduino.

Im considering a few different approaches, I could add a shift register (this is what i had originally figured that i would do). using 4 bits to source the current, and having the other 4 for added features, like buttons? or sound?, I dunno really.
Then I realized that mixing those bits in with the column pins may make shiftpwm, or other however I may control those bits more difficult, and a separate shift register may be better.
I figured, I could use 2 bits for sourcing the current per plane (1 8bit chip to source 1 48 LED plane), but it seems to me running the current for 48 LEDs through 2 shift register pins will be too much for it.

In my previous post, with the 4x4x4 RGB cube, they are able to do this with 64 RGB LEDs, and no other parts. I guess they are able to do all the current control via PWM, and since the red LEDs need a lot less voltage than the other LEDs, it somehow must adjust.

Also, Ive noticed other people posting about making 4x4x4 RGB cubes, if you are making a similar cube, post in this thread with how you are doing it, and what features you are thinking about.