I don't understand your use case much. Do you inject SecureFileBean anywhere or why is it annotated @Produces? I can't see anything like this in SecureFileSystemBean class. Anyway I guess it behaves like this because of SLSB pooling which was re-enabled in WF 10 AFAIK. See WildFly 10.0.0.CR4 Release Notes.
Thank you for the hint, I understand, that this shows the result of my bad design, on which you commented.
Evidently it worked by coincidence due to inactive pooling.
After activation of pooling, the SLSB was still active on the second call and still held a reference to the SFSB it had resolved before.
Thanks a lot,