I think it would be useful if you told about this in more detail. The globe is such a common suggestion for every game, I think a lot of people are curious to know the downsides of it. Obviously it requires reworking the engine on quite a deep level, but putting that aside (let's say you're coding a new engine from scratch for the sake of argument), what are some downsides?
How do you define borders with a globe? Do you use a 2d map? Doesn't that just move the projection problem down a level? Now you have a 2d image projected onto a globe, actually increasing the distortion (or creating massive waste in your image files if you distort the image file instead). If you use some kind of data file that is not laid out like a 2d image, or with vectors, then how do you manipulate it? Probably with a custom editor or tool, which you now have to develop and maintain. Of course if you do that now you don't get to use any of the VERY IMPORTANT image based masking and shader tricks that Paradox undoubtedly uses, at least not without a lot of extra leg work.
The best method I came up for predefined maps is using a Cube Map, normally used in reflections, to basically graft the faces of a cube onto a sphere, and base province delineations from the pixels of the faces of the cube. This allows you to partially treat the sphere as a series of 6 different 2d maps, and use a lot of the 2d techniques on them. Of course they break in a nightmarishly annoying way at the seams (edges) of the faces...
Theoretically you could export these to 2d images but they would be almost impossible to image edit like Paradox province maps because they are flipped and cut off in unintuitive ways.
With a globe the surface you are viewing is obviously spherical, this causes many problems with the plethora of graphical techniques used in games like these. Instead of using relatively simple 2d texturing techniques and shaders, you now have to develop an entire array of, some basically brand new, 3d spherical versions. Texturing a sphere is a very annoying problem (if you want pretty art textures you have to develop techniques with a coordinate system that can take 2d textures and graft them onto a spherical surface. This almost guarantees ugly distortions or weird texture flips which you can only minimize), and using dynamic shaders to create interesting effects is the same problem, exponentially more difficult. Seriously some of the new versions of existing techniques are many times harder to implement and also perform much worse without specialized optimizations, even then are never as fast as the simple 2d/flat 3d versions. A lot of times they look much worse too.
Some clever vector based graphics programming might solve some of these issues but that is way too gigabrain for me.
Like I said these are probably all solvable, or at least mitigatable, issues, especially for professional graphics programmers, which I am not. Still I hope you get an idea of why it might be a lot harder than you think.
For paradox they would probably have to overhaul their entire engine. This might cost them as much as developing an entire new game, and for a very unclear benefit for the games they currently produce. You don't need the poles in their current games.
