At the very least, this will apply a 'many eyes' to some of the SSL code and so long as the ssh pruning isn't wholesale slash-and-burn that cutting it back may prove efficacious for two reasons.
Less code can be simpler code, with decreased likelihood of there being a bug due to complexity and interaction.
Getting rid of the special cases such as VMS and Windows also reduces the complexity.
POSIX I'm not sure about; in many ways POSIX has become a dinosaur. Quite a number of Linux authors have observed that if you stop being anal about POSIX you can gt code that works and a simple #ifdef can take care of portability. In the 90% case there isn't a lot of divergence between the flavours and in the 99% case the #ifdef can take care of that.
Whether SSH fits into the 90% or the 99% I don't know. The APIs for 'random' and 'crypto' are in the grey areas where implementations differ but also one where POSIX seems to be the most anal and 'lowest common denominator'. I suspect that this is one where the #ifdef route will allow more effective implementations.
We shall see what emerges, but on the whole the BSD team have a reputation for good security practices so I'm hopeful about the quality.
I'd be interested to see their testing approach.