[MERGE] Faster commit

[Ian Clatworthy]

Ian Clatworthy wrote:

> John provided some alternatives w.r.t. locking directions. I think this
> is an important area for improvement given the coding tensions involved:
> low level implicit locking is safer and makes the API nicer *but* it can
> and does hurt us a lot performance wise. I've gone with selected low
> level lock removal and some better documentation for now. Is that
> enough??? Do tests need changing? (All current tests pass, of course.)

Looks like my testing was substandard - quite a few tests have indeed
broken since the implicit locking was removed. I'll fix this, take out
the message buffering (too contentious) and resubmit.

Ian C.