Downsampled stuff can sound much better if pre-treated with
signal processing techniques.
These would be dynamic compression and gentle high-end boost.
If you folks have analog compressor or comp-limiter,
and a parametric EQ, you can do this in analog.
(I don't have nice effect code yet for these.)

16-bits to 8-bits:
Dynamic compression pushes quieter signal detail from bits 8-10
up into bits 0-7, so that when you throw away bits 8-15, you
have a little more signal.

Lowering the sampling rate:
A gentle high-end boost will accentuate what high end you have
left after changing the sampling rate.  That is, by boosting
your "new" low high end, you imply the "old" higher high end
which is now gone.

Noise
talk about masking noise.