iirc, for the very project the decoding is done outside of Pd anyhow (on specialized hardware), so the problem of [unpackOSC] not knowing what to do with a blob is a minor issue here. (and the use of specialized hardware is the reason why data (low-resolution fixed-point numbers) should actually be packed in a blob rather than floating-point).
The decoding should be done on a microcontrollerboard and in puredata too. The audio data will be sent over ethernet to the target device, there it should be decoded and played back. The problem lies at the ethernet layer, where only a 10Mbit/s wire is available. At a Samplingrate of 44100Hz and a word width of 32bit are only about 7 channels (without overhead,...) possible. So it would be fine to reduce the resolution from 32bit floating point to maybe 16bit or even 8bit fixed point (while 32bit floating point is still selectable in case of better hardware). Sending 16bit or 8bit samples over OSC (without losing the gained bit width (OSC-data-types are at least 32bit)) will require an arbitrary alignment of the bits within each OSC-data-argument, where BLOB comes into play.
i still think (see above) that even [packOSC] cannot provide a consistent interface to pack blobs into an osc message that is sufficiently more simple to use than manually constructing OSC messages.
Thanks
Wolfgang