Par1,
In addition to what Greg Locock mentioned, I would start by testing each device by itself, if you can. Check the spectrums after you are done and see if you see anything that compares to the spectrums you have for testing with all devices present. If you cant find something there, test with one device, test with two devices, test with three, and so on. By changing one device at a time, you might be able to pinpoint what is causing a spike.
Are any natural frequencies being excited by the whole system being coupled? Do you have a soft tipped hammer to excite the structures yourself (without damaging anything) and get some natural frequencies?
Regards,
Fred