That seems like a ridiculous implementation of a quite reasonable algorithm. Instead of spreading them out through time, you can spread the values out through space. Assume you know that you elements are all positive integers less than 100 (or can be mapped to such). Allocate an array of length 100. In O(n) time, you can read through your unsorted list, and copy each element to the appropriate location in the array. Then you can read through the array and print out the values in order.
Similar but asymptotically different. In insertion sort, if your partialy sorted list is [1,5], and the next element to insert is a two, you would scan the list until you arrive at the five, which. In the system I described the partialy sorted list would be represented as [1,_,_,_,5], and you would update the second element so it would be [1,2,_,_,5].
EDIT: After thinking about it, you could do the insertion step in O(1) time if you use a linked list instead of an array. The part where insertion sort gets its log(n) factor is in scanning the list, which can take (on average) no less than log(n) time.
tantalor is correct in identifying this algorithm as pigeonhole sort.
