Mixing bit depths

The way I see it, the 16-bit samples would be converted to 24-bit, and then finally (in the mixdown) to 16-bit again.

No, converting 16-bit samples to 24-bit is just adding 8 zeros. This is perfectly lossless and doesn't damage anything. The processing --- fades, eq, reverb, etcs. --- that you do on the track are performed in 24-bit (although in SONAR they are actually performed at 32-bit --- or 64-bit in SONAR 5's double-precision audio engine). In SONAR you don't have to convert the tracks first, the processing in SONAR automatically converts them internally. Later, when you mix down, the 24, 32, or 64-bit audio is dithered to 16-bits. This dither loses some information, but only below -96db so it is not audible.

Regards,
David