Currently, the number of instructions is always evenly split between the pseudoclocks. I don't think this has to be the case though, we could presumably store the location of the instruction boundaries so the instructions could be allocated unevenly between pseudoclocks. That would allow you to segment off a few slow devices onto a pseudoclock with, say, 1000 instructions, leaving the remaining instructions for another pseudoclock(s).
Currently, the number of instructions is always evenly split between the pseudoclocks. I don't think this has to be the case though, we could presumably store the location of the instruction boundaries so the instructions could be allocated unevenly between pseudoclocks. That would allow you to segment off a few slow devices onto a pseudoclock with, say, 1000 instructions, leaving the remaining instructions for another pseudoclock(s).