looking at your code, the thread is joined with the main thread after the draw call. What synchronization do you have in place to make sure the buffer is not being modified during the draw call ?
Using fences/sync I've been able to upload in one thread and use the data for drawing in the main up to and including the latest Catalyst drivers. Unless you can guarantee that the upload if completed before actually using then the result is undefined. Unless the buffer is mapped with the persistent bit set, then what you are doing should come up as an invalid operation.