> On the other hand, ordinal systems have to be much more coordinated as they are not trivial to parallelize

I don't think this is true for all ordinal systems is it? Wouldn't most condorcet methods simply be tallied into a head-to-head table at each precinct and then combined centrally? It's mostly just IRV that has this issue since it's just iterated first past the post, and the order of elimination is important to know in each round before it can be continued. When all that's needed is a h2h table (no iteration) this isn't an issue.