Wow, errors of that magnitude should be telling you something is terribly wrong with the measurement! Keep in mind that this is a dB scale...if you're off by 3dB, then you're off by 200% - just imagine a volt meter that reads 6V when it's really 12V.
Errors under 1dB would be acceptable...maybe a bit more if you can verify environmental noise is tainting the measurement, but if noise is the problem, then the measurement should most certainly NOT be used for any form of decision making about the system. In other words, you should be ignoring the regions where the measurement isn't valid.