As someone who writes and supports profiling software the simple answer as to why different software gives different results is that each piece of software uses different algorithms to generate the profiles. The biggest variables are likely in the regression algorithm that is used to fit the LUT curves to the data points pulled from the target. There are a whole range of algorithms that are used. These vary in complexity, speed, curve smoothness and other factors.
The software designers who work on this are working with a difficult problem in that they are trying to create smooth curves that fit a noisy sparse data set that extends well above and below any of the known data points in that data set. As a result there are always significant trade offs that are made. Do you favor smoothness over tight fit? Or is speed more important? An so on. Each software designer makes different choices. As a result you as the user will get different results at least in some specific cases.
I have been working with Wolf Faust (for those that don’t know Wolf is one of the larger producers of IT8 targets and he also produces OEM targets for some well know IT8 vendors) on some profiler tests using a ultra wide gamut set of 5 target slides. One of the targets is a standard IT8 target with the patches pushed to the full gamut of the film and the other 4 contain additional patches that cover the full gamut of the film he is using (new Velva) for a total of 1440 patches. All of the slides are hand measured.
The initial tests indicate that some scanners have a hard time handling this much gamut. For example some Nikon scanners fall apart on highly saturated greens. And that some profiling software falls apart when feed images with this much gamut at least from some scanners. One commercial package which ships with a widely used spectrophotometer from a major US manufacturer has its wheels come off with highly saturated greens for example. Both open source profilers, LProf and ArgyllCMS, handle these same images with ease and generate profiles with nice smooth LUT curves that do not show any anomalies.
Wolf will be making the sample scans from these slides available to the public for wider testing the a few weeks. In addition anyone with a high end scanner who is interesting in providing a set of scans for this effort should contact Wolf. If he does not have any scans from the type of scanner you have then I am sure he will want to work with you. Eventually we hope to have a wide range of scanners represented and also to have profiles from a wide range of profiling software from these scans so that the different profilers can be systematically compared. My gut tells me that some of the results will be surprising.