If we design subroutine call such that we see the return address as the first operand of the callee function (because it is very easy to calculate the offset of return value), we will have some stack manipulations to bring effective operands (actual operands of the callee function) to top of stack to compute. These are some hidden overheads if the compiler is implemented that way. This EIP will overall reduces the pressure on the stack (by removing one element from it). The scheduler should be capable to generate better instruction sequences by reducing the stack manipulation opcodes needed.
Or, if you implement in such a way that you see the return address value as the last operand in the function, then this EIP’s performance impact will be reduced.