Actually I think this goes to architechture. I suspect the lock is less important on a single HT processor that has a uniform view of memory, either via the cache or write buffers. I'd still use them. My preference would be to serialize access to structures rather than do multiple locked RMW actions on elements within the structure, but to each their own.

If there is a chance that the instruction flow might not be serialized then use the lock. I'm not sure exactly how HT dispatches instructions and does load balancing, it could fill executions units in parallel from two or more instruction streams in which case the lock would be needed.

I guess I'd like to know why you don't want to use locks? Multithreading is problematic, any dual processor system (physical/virtual) will likely catch race conditions you wouldn't see on a single processor so you need to program defensively because you ultimately can't control the environment your code might actually run on, now or in the future.

If you don't want to use the lock because it costs too much, then you might want to revisit your algorithm design to eliminate or reduce your need for them.