The only thing that I am aware of with Pd using multiple threads is when used with JACK on Linux, as Jack's design requires a separate processing thread. Not sure about other areas of Pd though.
i do use pd+jack+linux for a while now because jack offers great possibilities, but i guess low-latency realtime audio always has to be seperated from other functionalities to work smoothly.
i think that pd should be able to load a simple soundfile into an array without blocking for ages. also i think it should be possible to draw a line in a graph/array without making the whole application stutter.
there are workarounds and one can learn to avoid situations, but ideal would be a pd taking care of such things.
regards d13b