I'm quite a newbie here on the forums, but I've been looking to both modelica and Xcos and here are my impressions. Anyone can feel free to correct any mistake I might write.

Xcos follows a so-called causal approach, in which the model is built with a flow of information following the solution procedure
Modelica foollows an a-causal apprach, in which the model is based on equations, not taking care of the solution of the problem

Trying to give an alternative explanation, the main outcome is that when you build a model in xcos, you have to know what are the variables and unknowns of your system of equations and you have to build a system in which information floew from the variables towards the unknown (so you have to build the solutions system)
Whenever, even in the same system, you swap a the unknown with a variable, you have to completely or partially rebuild your system.
In modelica you just use the equations, so you can build a system in an easier way, you can easily swap variables and unknowns and you can easily reuse your system or parts of it. Of course this tend to depart from the solution itself of the problem, it will be harder (at least) to know the process used for solving the problem.

Regarding compatibility, I saw that Xcos has some modelica library, but I never dared to use them, as I'm not so developed human being. I don't know anything regadin reverse compatibility.
Googling around I found this document that could be maybe helping.
http://www.openeering.com/sites/default … delica.pdf

That's all I know about it. I hope it helps.
Cheers