=========================================================== autovacuum_vacuum_cost_limit (integer)
Specifies the cost limit value that will be used in automatic VACUUM operations. If -1 is specified (which is the default), the regular vacuum_cost_limit value will be used. Note that the value is distributed proportionally among the running autovacuum workers, if there is more than one, so that the sum of the limits for each worker does not exceed the value of this variable. This parameter can only be set in the postgresql.conf file or on the server command line; but the setting can be overridden for individual tables by changing table storage parameters.
So, that's pretty clear, I think. In practice the balancing happens in autovac_balance_cost() function. The autovacuum workers communicate through a chunk of shared memory, and cost rebalancing is one of the things doing that.
Regarding the limits - yes, the 8MB/s is based on vacuum_cost_page_miss=10, which means a read from the OS. Per second, there's the worker process wakes up every 20ms, so 50x per second. As each round has 200 tokens, this means 10.000 tokens per second. Assuming all of them are reads from disk/OS, we can do 1000 of them (because the cost is 10), Which is 8kB x 1000 = 8MB/s reads. OTOH writes are about twice as expensive, leaving us only 500 writes, i.e. 4MB/s.