Fairly safe to say that they both 'hand off' the cpu for the duration. The benchmark as run proves little else of course. I'll leave comparing their efficiency/accuracy AAEFTR.

#! perl -w
use strict;
use Win32;
use Benchmark;

timethese (  100, {
    'Win32'    => ' sleep 1',
    'Std'    => ' Win32::Sleep 1000' ,
} );
__DATA__
# output
C:\test>183845
Benchmark: timing 100 iterations of Std, Win32...
       Std: 100 wallclock secs ( 0.00 usr +  0.00 sys =  0.00 CPU)
            (warning: too few iterations for a reliable count)
     Win32: 100 wallclock secs ( 0.00 usr +  0.00 sys =  0.00 CPU)
            (warning: too few iterations for a reliable count)
[download]

A quick 'deco' at the task manager whilst either sleep is in 'progress' shows that they are fairly obviously doing 'nothing'! :^)