Sorry for replying just now not knowing the question was posted 14 years ago when hardware technology was not as advanced as now (2025).
But for helping others improve their fast sorts (quicksort, shellsort, mergesort and others), I would post the results that I achieved yesterday ( 29Oct2025 ). Sort done using Shellsort compiled using Microsoft Visual Studio's C++ that was executed on an Intel i7-1355U CPU ( mobile computer CPU that draws only 15w of electricity ) with 16GB of memory. Base CPU speed is 1.7 Ghz.
20.733 secs for 200 million records
9.007 secs for 100 million records
0.87 secs for 10 million recods
0.078 secs for 1 million records