And here's the next thing:

That's two NB3N551 clock fanout buffer chips. The bottom one is being fed by a 20 MHz oscillator and the wire leaving from pin 2 is around 4 feet long before coming back to pin 1 of the top one. By the time it gets there, on the scope, it's almost a triangle wave. But after exiting pin 4 and going over to the ATMega, it's just as pristine as the square wave leaving the oscillator. The first stage of the NB3N551 is, so far as I can tell, a very, very fast comparator comparing the input to Vcc/2. You should be able to feed it just about anything and it'll turn it into a square wave, as long as it is over the Vcc/2 threshold 50% of the time. ¡Me gusta!

Now, this circuit doesn't have any bypass caps on it and there's no termination on any of the clock lines. So it's working more or less perfectly under definitely the worst case conditions.

So I feel good!