I would expect a kicker to have slightly better performance, since it is builtin to the system and does not have to communicate over the API.
Generally, you can think something like this:
If all you care about is triggering an action (such as re-deploy), then a kicker is generally the right answer.
If you want to perform some other procedure (say that you want to copy some data around, send a notification or similiar) then a subscriber makes more sense.
When using a kicker, almost all the logic is in the triggered action, but when using a subscriber you can have more logic in the subscriber itself.
