This is simply down to the nature of an optical compressor, when the unit has experienced no audio for more than a few seconds it passes full level. The initial transient will be louder, following transient of the same amplitude will be at a lower level as the optical sensor takes time to recover after a loud transient (slow release). 

An optical compressor can never be a fast attack compressor when compared to FET or VCA compressors.  

For this reason many people use a faster compressor (e.g 76-KT) in series with the optical compressor to get the best of both worlds.