" Is the hardware divider pipelined, is there more than one per core?"

Chaining the divisions (as a series of dependencies) would enable one to see the full latency of a single divide. You could use this data to estimate the number of divide units on the core.