I was just experimenting with core.atomic and noticed only a marginal speed difference when accessing variables using atomicLoad. Is this because it isn't implemented or because it can be reduced to a no-op on x86 hardware? -- Marco