I tried copying the calibration point list to my Cal file as suggested by tv84, I also added in some of the missing data of the following two blocks that appear to have very similar structure. The result was worse than with the unmodified file, with a very distinct amplitude jump (drop) at exactly 25MHz (i.e. when stepping from 25 MHz to 25.000 001 MHz. Moreover, the amplitude droop towards 100MHz was as pronounced as when I ran the generator without any calibration file, i.e. it dropped by half. With the original ("incomplete" -> 30MHz) cal file, it's much better and droops only by round about 2.5dB, while with the DG2000 cal file, it doesn't droop at all vs. frequency but has other inaccuracies (DC offset, general amplitude and frequency figures).
Edit: The divisor for the cal frequency list appears to be 3906.25, resulting in cal points (among others) at exactly 30MHz and 25MHz, which my test described above suggests. Since by replacing the cal point list of my original DG800 file with the DG2000 file, the cal points got shifted down by 16 places, the 25MHz calibration will now point to a place that may not be specified (even though I filled in some data from the DG2000 file that I assumed to be cal parameters), resulting in the "jump". This means, there has to be a "hard" correlation of cal file locations regarding frequency / parameter allocation.
The "second" calibration frequency table, located at 0x438f, starts in the DG800 file at 500kHz and as the last two frequency points, holds 30MHz and 100MHz, whereas the DG2000's file starts at 300kHz and has the additional "Low End" cal points 600kHz and 900kHz. From 1.25MHz to 30MHz, the cal points match between the two while between 30MHz and 100MHz, the DG2000 cal file contains 11 more points. This quantity of cal points and also the presence of more low-end cal points on the DG2000 makes me assume that this instrument is calibrated to a higher accuracy vs. the DG800 / DG900(?) series, legitimating its "higher rank" in Rigol's AWG product range. Maybe the customers are paying for calibration accuracy rather than for hardware value. Just a wild guess, though...
The blocks following the Cal Points, starting in 0x451e and 0x45e6 appear to hold the calibration constants, consisting of little endian DWords. The number of entries matches the number of Cal Points and apparently, the value of the constants is continuously rising (which makes sense as when the frequency increases, the gain drops). The presence of two such cal blocks for each channel makes me assume that these two blocks calibrate the two output amp configurations. I've now got a pretty good idea of what's going on there and could possibly fit a curve on the DG2000 cal points to extrapolate the existing cal point ensemble of my DG800 in order to generate the missing ones between 30 and 100MHz. If I find some time, I'll give it a try...
Edit 2-3/8: I started analyzing the cal data and it appears more complex than anticipated. The first blocks of cal data don't appear to make much sense, neither in Integer32 nor in Float32 format. The second blocks may very well be cal factors in Float32 format. But maybe, the two cal blocks are meant to be used in combination, either to describe a starting value and a (linear) slope or the like, or maybe as a quotient or similar to improve resolution. I attached an oversize PDF and a zipped Excel file (converted form OpenOffice) of the lists. Maybe someone is interested in having a go at the puzzle...
Edit 3-1/4: Unfortunately, my recent experiments prove that only patching the cal file locations that I analyzed in my PDF, won't suffice to "smooth out" the upper frequency range. I copied the upper end cal data of the DG2000 that better matches my existing DG800 cal file, taking good care that there's no improper interval specification, and the result is worse than with the original cal files. The AWG will perform flawlessly up to 40MHz and show a sudden level drop there by about 10% (when crossing the 40MHz -> 40.000001MHz border), drooping further when frequency is increased. This droop is once again way worse than with the original cal file. The only explanation for this is that calibration information in other locations of the file is missing for a complete upper end calibration. With this information, it may be advisable to look again more closely at a proper SCPI calibration... Bummer!