I have recently released v1.0 of this library, based on a new algorithm. While the previous version ("production" release) is still valid, I was not fully satisfied so it wasn't given a version number. This version uses a true LRU, but doesn't suffer the lock contention that a naive implementation would have. It should perform better than the prior release and not rely on a pseudo-LRU policy for performance (which can degrade under certain workloads).
Please consider upgrading when convenient.