Using the following sequence, using REPL_SYNC:

[thread-A] nodeOne.put(K1, V1);
[thread-B] nodeTwo.startBatch();
[thread-B] nodeTwo.remove(K1);
[thread-A] nodeOne.withFlags(Flag.SKIP_LOCKING).remove(K1);

At the last line, thread-B is blocked waiting to end the batch to
after thread-A finished. This results in thread-A to throw a

"org.infinispan.util.concurrent.TimeoutException: Unable to acquire
lock after [500 milliseconds] on key"

Still if I run the same sequence of operation on two different threads
on the same node, this will work just fine and not throw any error:

[thread-A] nodeOne.put(K1, V1);
[thread-B] nodeOne.startBatch();
[thread-B] nodeOne.remove(K1);
[thread-A] nodeOne.withFlags(Flag.SKIP_LOCKING).remove(K1);

Looking into org.infinispan.interceptors.ReplicationInterceptor.handleCrudMethod(InvocationContext,
WriteCommand), it seems that any Flag is not sent to remove nodes.
Actually the whole org.infinispan.remoting.rpc.RpcManager interface is
not interested in Flags.

Test being send soon to GitHub.