In theory you could do this in hardware. Hardware timers are pretty trivial, you could imagine a hardware chip that accepts a number, then writes the value to the next free slot in an array when the time has elapsed. As timers go off, the array fills up. Obviously you need to handle collisions and whatnot but nothing in this screams that you need to go O(n^2).
Response to dead comment:
> Couldn't this be solved by scaling everything by the largest entry. Of course that requires first a pass to find the largest entry but that's not as expensive?
One presumes that we don't have infinite precision timers, so you should only scale it down so the smallest gap between elements is a few clock cycles (unless you're satisfied with breaking the fiction that sleepsort is not heapsort). This seems like a mildly interesting question. My intuition is that finding the smallest abs(a[i]-a[j]) with i != j should be just as hard as sorting the list on a classical computer, but it seems like a quantum computer might find the smallest gap faster.
O(n) sort achieved?