Jerry Coffin's answer was excellent, and I can't add a whole lot more. One other possibility may be the calibration hardware itself. There are two fundamental types of calibration devices: Colorimiters and Spectrophotometers. A Colorimiter is a "scientifically subjective" device. It uses an approach to calibration called "tristimulus", and is designed to mimic how the human eye perceives color. The human eye's cone cells are sensitive to specific peaks of color (blue, green, and two peaks of red), which when interpreted by the brain are combined to produce the range of color we can see.
Most calibration devices are colorimiters, and while for any particular screen they will generally produce fairly consistent results, they are not guaranteed to always calibrate exactly the same, and will not necessarily calibrate each device the same. This is generally not a problem, as they work within the bounds of human perception (and in the same way as human perception...hence their slightly subjective results). Their slight variation is rarely detectable except when recalibrating and comparing before/after results. Combined with different kinds of back-lighting such as fluorescent tubes (which tend to oscillate at a multiple of the frequency of their power supply), and you have a system with a few variables that can cause different calibration results for different devices/different times. Some calibration systems, such as the Spyder3Elite v4.0, allow you to synchronize the calibration of multiple devices by creating a shareable profile that may be used to normalize the calibration of multiple, disparate devices to a common baseline.
Higher end (and usually more expensive) calibration devices are often spectrophotometers (usually in the form of a spectrocolorimiter.) A spectrophotometers is a truly scientific device, and is able to generate consistently accurate measurements at all times. Spectrophotometers can still be susceptible to lighting oscillations of fluorescent tubes, however most calibration devices that use spectrophotometry (and even most colorimiters these days) take that into account and are still able to produce consistent results. Spectrophotometers use mathematics to convert their light measurements into estimated tristimulus values. Since the final calibration results are estimated via mathematics, rather than directly measured, while calibration results may be scientifically true to the measured light, they often have a slight, but consistent, deviation from what you would expect. Again, these deviations are usually not detectable by the human eye when not comparing images from before/after calibration.
If you suspect that your calibration results either being skewed by the backlight oscillations of your screen, or by inaccuracies of the calibration device, you might want to look into a system that is designed to calibrate multiple devices to a consistent baseline. The DataColor Spyter3Elite with the v4.0 software offers this capability. It is able to generate a sharable calibration target that can be used to calibrate any number of devices, of varying types (LCD, CRT, TV, laptops, projectors, etc.) to produce consistent output. Calibration will obviously take longer when calibrating multiple devices, and you may need to do a couple re-calibrations of each device to get perfect normalization, but if you do regular multimonitor work and need accurate calibration across all of your displays, the investment in a system that can calibrate them all is worth it.