How's this for a solution?
You could implement, say, 60 LED's driven with an octet of shift registers. Drive 8 shift registers with 8 output pins apiece and you can control 60 (64, actually) LED's. Implementing brightness and the fading (where certain LED's in the center are brighter than their counterparts on the sides) is just a matter of PWM. I'm not sure how many pins that would require. It's either 16 (one serial and one clock) for each, or 9 (one serial each and one common clock.) I don't know if you can get away with one common clock.
I suppose if PWM over serial is too clunky you could implement it using an 8x8 charlieplex instead, but the charlieplex would require 16 pins as well.
Once you establish control over the 60 LED's, the rest of the effort is centered around
a) implementing a clock, and
b) choosing the presentation.
That is to say, choosing whether you want the blue or the red LED's, whether you want a dark background with bright "hands" or a bright background with dark "hands," etc... Bookkeeping entries, that aren't too terribly interesting, though will probably cost 10x as much time in the debugging.
I suppose if you want multiple colors you're multiplying the number of pins you need by another ugly number. RGB LED's mean you need to control 180 cathodes instead of 60 which probably means you've no bloody choice than to cascade multiple shift registers into each other. The LadyAda comments in the partselector page suggest that's possible but as for pulling that off: banned if I know.
Red M&M, Blue M&M: They all wind up the same color