I don't know on a provincial level but maybe on a per-state level, since if the national pool is gone, then that flat 10000 bonus (that makes small nations viable) no longer makes sense so you need a geographic unit large enough that you can recruit more than one unit every few years. If we give states manpower pools, then the religion and culture of a unit could be a percentage chance weighted by the manpower of a given culture and religion in that state (thus, the only way to predict 100% what a unit would have would be to have an entire state be uniform culture and religion, then the only way to build a sizable army with 100% prediction would be to have a lot of states like that; any decently large empire will have to diversify their army). The only thing I wonder is that would result in divisions that could theoretically not line up with a combination even present (say...English Protestants and Welsh Catholics in one state could theoretically make English Catholics or Welsh Protestants, though this is easy to justify if you point out that "religion" just means "dominant religion")
I assume by loyalty you mean that when separatists or zealots rise up, your units of that culture/religion rise up with them? In this rework, would this include professional units? Conscripts of course should rise up, but professionals is another matter. And it could be good stepping stone if you eventually want to build towards a real civil war mechanics, where different factions in the army align themselves with different pretenders based on culture, religion, traits (more tolerant rulers would get more support from non-primary culture units), etc.
Edit: that being said, I don't know that that's enough to justify it as is. I think the idea has potential but if you are going to add this degree of complexity you have to have a purpose behind it, and I'm not sure that's enough on its own. But there will probably be potential to the idea beyond what I can think of. I hope other people can think of ideas to use it beyond this.