@Nikita: The user cache supports concurrent builds by design and we test for it. As evidenced by this thread, there are some issues though. So far we’ve been unable to replicate the scenario that causes this problem to surface in this way.
We do know that under high contention across long running builds, this problem can occur. What we don’t know is exactly what the lock holding build is doing that prevents it from relinquishing the lock for another build. That is, some operation is locking to coarsely and is not respecting requests to relinquish.
Your options are to lobby for a fix, submit a fix, or distribute builds across different user level cache locations.