Recently the new charge injection calibration data was collected. I have compared 2010 CI calibration with one collected on orbit in 2009 and on the ground in 2008 (which is actually used for data processing) .
The difference in ADC response between 2010 and 2009 calibrations as a function of injected charge in DAC units is shown on the following plot for "typical" channel twr=0, lyr=0, col=0, face0 in LEX1 range (this range has the biggest deviations from linearity and for it the charge injection calibration is the most important):
Similar difference for the same channel between 2009 (orbit) and 2008 (ground) is following:
In both cases the difference between two charge injection calibrations is mostly linear and is ~0.2% for 2009 - ground difference and several times less for 2010-2009 difference.
The histograms of linear slopes for all 3072 LEX1 channels is show on next two plots for 2009-ground and 2010-2009 comparison, respectively:
For 2009-ground comparison the slope is almost always negative and less than 0.6% (there one outlier with positive slope). For 2010-2009 the difference is smaller by order of magnitude.
IN fact, the effect of additional linear slope in charge injection is taken into account by MeVperDAC calibration, so it is given here just to show the scale of variations in charge injection calibrations.
Important is the deviation from linear fit, show for all channels on two following 2D plots: 
Horisontal axis gives log10 of DAC values, while the vertical axis is in linear scale and gives difference in ADC response.
In 2010-2009 comparison the deviations from linear fit are negligible, for 2009-ground comparison they are clearly visible, but rather small (~0.1%).