In my (2 cent) experience to let the un-processed signal go _through Pd_ is still unsatisfying, because anyway you have to deal with some latency hiding around.
Instead I usually let my external soundcard output a small amount of "direct monitor". This way my clean signal goes through the hardware to the PA with no latency (or at least a fairly un-noticeable latency), then the processed signal/s get overlapped with the minimum latency I can get.
This is the cheapest and more satisfying workaround I found so far, even though it's not reliable in every condition and setup.

Besides, as already said, when you trigger digital effects with an external hardware you can add more latency; to avoid it, I analyze the un-processed input signal, recognize the notes being played and trigger effects when a note is recognized.
Thus I just deal with the latency created while the un-processed input reach the analysis patch.
At the same time I route the un-processed signal to the effects bank.
Not rocket science but hope this helps.


M