I agree with Ajay Shukla, it basically depends on the publisher, the scope, the nature and other necessary parameters set by the organization to make them unique and at the same time, produce a high quality, high impact research. Basically, similarity index between 10-20% is substantially an acceptable range.
Yes, as most of the oath said, this actually depends on the journal. This number even could be changed from one year to another based on the managing behaviour, number of submissions, number of quality papers, etc. However, I agree that usually journals follow an acceptance rate, may be as a parameter to control the quality of their output. A study conducted the journals indexed in DOAJ (within year 2017) reveals that the majority of journals in medicine and social sciences have average acceptance rates (41%-60%). The least percentages were reported by the journals with very low acceptance rates (1%-20%). These findings were common for both subject domains. It is obvious that most renowned/reputed journals belong to the category that have very low acceptance rates, yet one can still find reputed journals with relatively higher acceptance rates. Therefore, according to my opinion, selecting a journal for publication with a certain acceptance rate depends on the quality of the paper you produce.
It depends on the journal and what is considered. Exlcuding the reference, it's 10% or less. However, many mathematic concepts and variables are the same. This raises some problems.