Nevermind, there is no regression vs. 1.63 here, the problem was in my user-config.jam file (how many times has that happened!). 

For both the python 2 and 3 configs, I had a 32-bit rule that included the conditions section:

: <address-model>32 <address-model> # conditions 

The goal here was to support both builds with address-model=32 and no address-model specifier. On 1.64.0.beta.2 this causes it to not find any python version and fail silently.

On 1.63 it shows up as an error because python couldn't detect a suitable version.

For my build script I simply removed the conditions, so if the address-model isn't "64" then it will use that version...this works, but I had be lead to believe that the original line would work as well. Another thing that *doesn't* work is having two separate configs for the same version (for a total of three when including 64 bit) with one specifying <address-model>32 and another with no conditions. This caused an immediate error that prevented the build.

This clearly isn't a show-stopper, but it seems like it should be cleaned up....not sure how.