I think you're going to need to perform experiments and decide if the flexibility you get is worth the performance you're going to lose.
As you can see from John R Pierce's response it is only matter of time before someone (correctly) warns you about the performance issues with EAV derivatives, but then the conversation becomes too abstract, at least in my opinion.
Why? Because we don't know how much data you're going to process with this design, with what kind of hardware.
We also don't know how the data will be processed.
Where EAV dies a screaming horrible death is when you try to search on multiple attributes. If this data is just never going to be searched on attributes, EAV isn't that bad (I use it for some things where that is clearly the case).
