This was recently discussed on the nuke-dev mailing list and some approaches were new to me, so I thought I would add all answers here:
* Use the -P switch to pute nuke into performance mode
* Have a test .nk (e.g Read->MyNode->Write) and compare times how long it takes to render on the command line with "nuke -x"
* set the number of threads to 1 (nuke.env['threads']=1 or -m switch) and put print statements in the engine function, with timing info (e.g time elapsed from start of method to end)
* use valgrind's callgrind (using kcachegrind or something to view the results) - valgrind slows everything down, but the relative times should still be valid
Thanks to all the helpful people on the mailing list