Currently, 349K wps is the worse case on a RAM disk, 291K wps on a 7,200 rpm hard disk, and 221K wps on a 5,400 rpm disk. (Note: I didn't have a SSD on the test system.)

Not to detract from your code -- if it achieves your goals, its perfect -- but by way of comparison with using real shared ram.

The following does 32 million fully interlocked writes (32 threads x 1e6) to real shared memory in 4.17 seconds (7.6M wps):

C:\test\lockfree>type lockfree.c
#include <windows.h>
#include <stdio.h>
#include <time.h>
#include <process.h>

typedef struct {
        int i;
        int loops;
} shared;

void worker( void *arg ) {
    shared *s = (shared*)arg;
    int i = 0;

    for( i=0; i < s->loops; ++i ) {
        _InterlockedIncrement( &s->i );
    }
    return;
}

void main( int argc, char **argv ) {
    int i = 0, nThreads = 4;
    clock_t start, finish;
    double elapsed;
    uintptr_t threads[32];
    shared s = { 0, 1000000 };;

    if( argc > 1 ) nThreads = atol( argv[1] );
    if( argc > 2 ) s.loops = atol( argv[2] );

    printf( "threads:%d loops:%d\n", nThreads, s.loops );

    start = clock();
    for( i=0; i < nThreads; ++i )
        threads[ i ] = _beginthread( &worker, 0, &s );

    WaitForMultipleObjects( nThreads, (HANDLE*)&threads, 1, INFINITE )
+;
    finish = clock();

    elapsed = (double)(finish - start) / CLOCKS_PER_SEC;

    printf( "count: %lu time:%.6f\n", s.i, elapsed );
}
[download]

A run:

C:\test\lockfree>lockfree 32 1000000
threads:32 loops:1000000
count: 32000000 time:4.171000
[download]