Facing this same problem, I installed the java sources with the SDK and then selectively imported subsets. For starters I imported the entire java. namespace and then the entire javax. namespace that comes with the SDK. (Note that this is not the entire javax namespace, just that subset in the SDK.)
At this point the project file was approximately 38 MB in size. I then created a master, for master/replica creation, and that expanded the baseline project file to about 51 MB before we began bringing in any of our own code.
The advantage to this approach is that everything is available to me as a designer from Swing to java.sql to java.util.regex. The disadvantage is the overall size of the master project file. If someone would like a copy, I can see about some way to make it available. The java/javax baseline, without conversion to a "master" is 38MB and compresses to 11MB.
I do not have the ability to email 11MB images nor do I have a place to host such an image for FTP access but I am more than willing to make it available to someone if they are interested. Or, if Sparx would like to host it, perhaps some solution could be worked out there.
Note that I had to resolve about 2-3 dozen ambiguities during reverse engineering and in many of these cases I guessed rather than opening yet another source file and trying to figure out what the tool was concerned about. I have not yet found that any of these guesses are "wrong" nor have they impacted my own design efforts yet and if they do, I will simply amend my project as needed.
The entire importation process took me the better part of an afternoon but I was also doing other activities during this time frame.
I give high marks to EA and its reverse engineering tools for scaling to such a huge base and still maintaining a small memory footprint, fast speed, and so far, great stability even still.
