A network engineer spotted some packet loss averaging 4 packets in 1000 then some body tried to use that as a cause for a performance issue. Is this a realistic assumption that that sought of loss can cause application performance issues.
0.4% packet loss is quite high on an internal network. Packet loss will definitely affect performance, but the effects are also dependent on RTT.
The Mathis Equation describes the maximum TCP theoretical throughput for a given RTT and Loss Rate:
MaxRate < (MSS/RTT)*(1 / sqrt(p))
Where p is the probability of packet loss.
For example, with a RTT of 50mS and a probability of 1E-4, you couldn't go faster than 230Mb (assuming an MSS of 1450).
Here's a short explanation.
(Assuming TCP over public WAN)
Yes - your equipment is taking a performance hit when packets get dropped because data packets are being resent in response to the ones that got dropped.
A 0.4% packet loss rounds to zero for a reason. It's not bad for internet traffic quite honestly and many professional network engineers might even say it's an acceptable drop rate for a wired internal network. My opinion is that if you were able to instantly correct the problem and then use the application in a state of zero packet loss, it would be next to impossible to tell a difference.