Steve, my code does exactly what I want... it gives full time low output when set 0 and full time high output when set 255.

I think you are getting a bit hard with me.
Of course cleaness is just stetic, but it is very clean to me. Gives 3 or 4 cycles per PWM service.
PWM value of each output is never modified, only the PWM "position" or "cursor" (dont know the right name).

I just thought my code was a bit slow, giving just 205 refreshes per sec, but appears this is not too bad.

(4 * 8 * 256) + (256 * 4) = 9216
1000000 / 9216 = 108,50
I think this calculation is not 100% right, but is not completly wrong

Switching to another uC is an idea, of course.

I can put my code in a timer interrupt, I think there is no problem, but, as I understand, this will reduce the refresh rate per sec.