There are a few variables to this question including what OS you're running and what specific AV you're using, but I think it's fair to say that exclusions don't necessarily eliminate the impact of AV. These products require some sort of file system filter that allows them to intercept events so there's always the potential, however small, for problems.
I ran an experiment with an AV product on Linux (CentOS 7) to see what kind of impact it would have on disk performance. In my experience the biggest impact with AV is observed when creating files so I used bonnie++ to measure the latency involved when creating thousands of files. With the AV service stopped, the latency was around 100 microseconds and appeared to be CPU-bound on the testing process. As expected, without any exclusions on the target path the latency was so bad I couldn't wait for the test to finish. The particular product I tested is interesting because it has two different interception methods available: a vendor-provided kernel module and fanotify (which appears to be a generic interface provided by the Linux kernel itself). I tested an exclusion with the vendor-provided module and the performance was outstanding, as though the AV wasn't there at all. On the other hand, with fanotify the latency was about 200 microseconds and was accompanied by a lot of CPU usage on the AV scanner process. I thought it was an interesting test, that's just a 0.1ms difference in file creation time in an extreme test; is that likely to make a difference in any real-world scenario?
On Windows, my understanding is that AV typically uses file system minifilter drivers to intercept events. It looks like it's not unheard of for these drivers to cause unusual problems and I'm sure they could potentially impact performance. However, these products have been developed over many years and are used in production worldwide.
For what it's worth, we use AV with exclusions on Linux indexers and it doesn't worry me at all. The only extra I thing I do is check occasionally with a file creation script to ensure that the exclusions are still active; appending to a file a few thousand times and measuring the time taken is enough to confirm this. That said, comprehensive testing is the only way to be sure your performance isn't being affected. It might be time-consuming to test, but on the other hand if the results indicate that you should seek an exception to the policy, you'll have hard data to take to the appropriate group. Have fun!
... View more