I have the feeling you don't know what a Charpy impact test is or what it measures.
You DON'T measure acceleration anywhere in this test. You don't even care about acceleration. An accelerometer is possibly the worst sensor for this test.
The standard test method: a hammer on a pendulum is raised to a specific height. The test sample, of specific dimensions, is at the bottom of the swing. The hammer falls, hits (and hopefully breaks) the sample, then continues the swing. What you're then looking for is how high the hammer ends up. The higher the swing, the lower the energy absorbed by the sample as it breaks.
So to digitise the instrument, what you need to know is how far it swings, and that's exactly what an encoder tells you. Come to think of it, this must be a quadrature encoder so you can tell when the hammer starts to fall back again, which is when the measurement is complete. It's only the part where it swings forward the first time after the test starts that you have to measure.
Another possible method would a distance sensor that measures how high the pendulum swings. A fast measuring time of flight sensor would be suitable. The main problem of this method is that if the pendulum swings only a little bit, it may not come in the field of vision of the sensor. A second sensor (looking horizontal) would take care of that. Being contactless, this method will not affect the measurement but is definitely harder to implement.
As an example, your 400 count encoder is just going to be able to resolve a tad under 1degree rotation that’s if it can correctly count pulses at the speed max speed of the arc.
As long as you stay under a couple thousand pulses per second there should be no problem for the encoder, and no doubt there are high speed encoders that go way beyond that. Under a couple million a second will keep the Arduino happily counting all. I don't think that's a serious issue.