> Is the mbuf bounce buffer implementation nearly finished and we can
> commit it and do 3?

You said "mbuf bounce buffers".

The main purpose for this work isn't for mbufs, but to simplify the
buffer cache.

Improving the buffer cache situation will affect everyone.  Supplying
>3G of mbuf memory will affect very few people.

Maybe we should consider the buffer cache issues, which involves fewer
drivers, the priority.