It may be tempting to try to associate ERUF values with response times to estimate an ERUF "tolerance range" for your specific site.
Attempting such a process is very dangerous and can lead you to make serious mistakes that will cause many problems later.
For example, you may create several ERUF=4 functions and test them on a brand new (and very under utilized B60) to get an average response time of 1 second. Since you allow responses of up 2.5 seconds, you therefore conclude your "tolerance range" is from 1 to 10 ERUF units for any transaction.
However, the reality is that when the IBM i is heavily loaded with 50 online users and 7 batch jobs, the ratio changes and your ERUF=4 transactions are now averaging 3 seconds response.
Where does this leave your ERUF=10 transactions ? Probably with a response time of around 30 seconds.