Kip has pointed out that fakeit doesn't handle multiple sources correctly. If you set up with two sources with their own responses modelling a dataset then do a fakeit only the first source and response contribute to the simulated spectrum. This will take a bit of work to fix.
Jennifer Donley at Arizona e-mailed to worry about the warning message about incompatible CHANTYPE after simulating Chandra ACIS spectra for CY10. This is a minor bug in xspec which fails to pick up the CHANTYPE value from the response matrix when doing a "fakeit none".