Hi,
Yes if you have to divide the time up into intervals that are small you have to increase the peak current by the same factor, well that's if the LED had a linear current vs brightness curve, which it doesnt. It usually takes more current than the simple division, but even using the simple division and multiplication method for a linear LED we'd run into the problem of peak current which the LED probably is not even rated for even if you had a big transistor to drive it.
If the uC can turn on up to 10 LEDs during any ONE time segment, then the required current is 10 times the current of one LED. If one LED looks bright with 10ma, then we'd have to have 100ma capability. But that's assuming that the time segment duty cycle is 100 percent, which it never is.
So if we had 20 rows of 10 LEDs where we turn on 10 during any given time segment, that means the average current to one LED decreases by 20 times, or 1/20 of the current we had with 100 percent duty cycle. So to get the same brightness we'd have to pump the peak current up by a factor of 20, which brings us to 2 amperes. Each LED would be getting 200ma peak, which is already too much for the little guys (assuming a 20ma LED). Going to 40 LED rows we'd have to pump up to 4 amps, with 400ma peak in each LED. That's way too much unless we are using high power LEDs but then we'd want even more current for those.
So now you see the problem. The more scan lines to be done the less time each one can get. This leads to using more pins of the uC so that we can split the rows up into groups, or even going to what is sometimes called "CharliePlexing". That's where the LEDs can wired in anti parallel and driven using a somewhat different routine.
However, it's still going to take some higher than usual peak currents. This is where the extra pins come in.
You might be able to get away with a pin extender IC which is nothing more than a chip that is driven serially from the uC and provides programmable I/O pins. This effectively gives you a lot more io pins to work with. The circuit board layout wont be that much more complex however because the only input to this chip would be a clock and a data line which only means two uC pins.