Based on the title of this thread, the purpose of the dynamic analysis is for the seismic loading of the structure. I would, therefore, assume this is a response spectra analysis.
I'm not as familiar with the EuroCode specific requirements. But, my thoughts on the 90% mass participation:
1) Unless the structures are truly isolated from each other, I would include them in the same analysis. The point is you want to capture the interaction effects of the various portions of the structure. And, / or the various mode shapes of the structure. A simple (i.e. single mode) assumption doesn't work or else the equivalent lateral force / static loading would be sufficient.
2) The 90% limit isn't a magic number. 38 modes isn't that bad to reach 90% mass participation. If you want to do that in fewer modes, you might consider discretizing your weight / mass so that you avoid any local modes. All it's trying to do is sufficiently capture the behavior of structure. If you have capture all the non-rigid response in your 38 modes then you should be fine, whether or not you hit 90%. I consider anything above 25 Hz or so to be essentially rigid response.
The reason is that I'm assuming your scaling up your base shear to be some percentage of your static base shear. And, the rigid mode shear produces almost no overturning moment (because it's usually very low on the structure and close to the supports). Plus,it is usually associated with a spectral acceleration that is less than the other less rigid modes. Therefore, scaling up your base shear when this isn't captured is almost always conservative.
3) Many programs will have an option to consider "residual" modes or "missing" mass, or such. This is a good way to capture the missing mass participation (as long as it's rigid) without being as conservative as the scaling concepts in item 2.