I'll answer my own question here: BatchHttpRequest
is the way to go.
It provided only a small speed up over the multiprocessing code I'd already written, but the big benefit is that it simplifies the design a bit - especially important when we are bringing up thousands of nodes at once.