The jitter of the RC oscillator seems to be a better internal source of entropy than the ADC, though only generating 8 bits per second or 125 bits per second in example implementations.
For a specific processor, maybe. In the general sense, no. The AVR internal oscillator (and watchdog oscillator) certainly has jitter but the entropy in the jitter appears to be considerably less on newer processors. My suspicion is that the jitter for newer processors is primarily a function of the previous and current state of the processor (which is going to be rife with patterns) rather than thermal noise.