Yes, a study failing to replicate is not in itself evidence of fraud. Nor is it definitive evidence that the effect shown in the original doesn't exist, since a replication can be messed up too.
But science needs replication because we're ultimately trying to learn about the world, and an effect we can't reliably produce is usually an effect we're not sure exists. Replication sometimes catching dishonesty is a bonus: failure to replicate calls the effect into question, which calls the method used to find that effect into question, which sometimes calls into question whether that method was actually used. (Or in the case of bad statistics, whether it was chosen to dishonestly show a certain result.) It's the appropriate propagation of doubt.