Are you under Windows? Linux? Mac?
I encountered a similar problem myself. In my case it was because I installed numpy and scipy from the conda-forge channel, which does not install properly the mkl/blas dependencies. So, install the following from the default channel.
Try to create a clean conda environment
conda create -n env_name python=3 numpy scipy cython
source activate env_name
pip install git+https://github.com/theano/theano
Then open a python console and try to import theano only. Hopefully, you will get no warnings. If so, then install pyMC3
pip install git+https://github.com/pymc-devs/pymc3
Also, you could also check if mkl/blas has been properly installed by importing numpy and then type numpy.show_config() (if i remember correctly, i am using my phone now). You should get a list of configurations, and the name mkl/blas should appear several times in there.