One of ChucK’s raison d’etre is that it is "strongly timed". Considering the number of posts here about problems with Max’s timing (rather than MSP, of course), there’s a reason in itself to investigate ChucK.
It is a rapidly expanding language, with some very exciting recent additions. Many of them (analysis/resynthesis, for example) can be duplicated in Max/MSP with third party externals.
This summer Ge and company introduced sMIRk, a music information retrieval toolkit for ChucK. What that translates into for most of us is machine learning. The demo at ICMC and ISMIR involved giving it some audio data (piano music, then string music), giving it an opportunity to learn the difference (running the algorithm for a few seconds), then having it correctly identify new piano or string timbres, in realtime. It was a pretty sexy demo, to say the least.
This could possibly be done in MSP (using externals), but it’s ready to go in ChucK/sMIRk.
Other high level machine learning tools are promised, like HMMs: http://en.wikipedia.org/wiki/Hidden_Markov_model
Lastly, on the plane home from ISMIR, I was able to write a ChucK program (never having used ChucK before) that seamlessly talked to Max via OSC, using Max to send audio to ChucK, and getting (analysis) responses back, in realtime.
I am patiently waiting for Brad Garton’s chuck~ external for Max 5.
No pressure, though, Brad…