Hi,
just a tiny benchmark.
On my machine (P4-2400, Linux2.4.22) this 7076-type test,
compiled -O2:
for (int i = 0; i < 15000000; i++)
std::cout << std::setw(30) << i << '\n';
a.out > /dev/null
baseline
--------
9.820u 0.040s 0:09.99 98.6% 0+0k 0+0io 193pf+0w
__pad patch
-----------
8.790u 0.070s 0:08.99 98.5% 0+0k 0+0io 194pf+0w
Paolo.