So, what does InnoDB do while you’re loading the data in parallel?
It looks something like this:
Click on image for larger (8MB) version, if you dare, have efficient browser and lots of spare RAM (7k*10k canvas). Generated by:
opreport | Gprof2Dot | dot
domas mituzas 
Thats awesome!
Whaoh. I can see my house from up here!
Looks like at least 10% of the execution time is spent in trx_undo_seg_free(). You think this is where InnoDB is swapping the currently written log buffer to disk in order to make room for more transactions? Can increasing a buffer help reduce the time spent in this function?
-jay