It's not irrational. But, since you are willing to rely on C in addition to Python, why not code the parallelizable parts in C and leave the rest in Python.
And you can always write a proof-of-concept in Python, prove the algorithm is right and rewrite the performance-sensitive parts in C. If your tests are thorough, all is good.
That's unfortunately not an option because all requests would have to go through that single threaded bottleneck to get to the parallelized C code, which is only acceptable for long running operations. For a request/response type of scenario like Cassandra, Lucene or the kind of analytics systems I'm working on it would make the whole parallelization pointless.
The realistic option is to use a separate C based process behind the usual multi-process mod_python setup. That works but is more complicated and error prone than an in-process solution. So I think that's why so many new projects that have these kinds of scenarios (NoSQL for instance) go with Java in spite of its flaws.
And you can always write a proof-of-concept in Python, prove the algorithm is right and rewrite the performance-sensitive parts in C. If your tests are thorough, all is good.