What is the relationship between the Epoch, performance, gradient and time in the neural network training tool?

After training different set of data and recording some parameters and performance from the neural network training tool,i noticed that in some cases the network had a higher number of epochs say 1000, 300 or 200 and the performance and gradient were very good which translated to good results though it required longer time while in some cases the number of epochs was a bit less say 100 but the performance and gradient was not too good. How do i explain this trend? Thanks