Is there a way (in Blender) to limit the number of CPU Cores used for physics baking, like you can with rendering?
There isn't really a straight forward answer for this, it will depend on what simulation you want to bake and what OS you are using.
For simulations using openMP, setting the
OMP_NUM_THREADS environment variable should limit the cpus used.
For simulations using pthreads, adjusting system limits should be able to limit the cpus or cpu time used. The way to do this varies between most operating systems, for linux there are
/proc/... settings or the bash builtin
Another approach available on most *nix systems is the
renice commands to adjust the cpu priority so that other procs are given cpu time before blender.