[Haskell] performance tuning Data.FiniteMap

[Simon Peyton-Jones]

| But in managing this tradeoff, what is faster:
| * constructing/destructing e.g. 16 trees (for a 65000 item table)
| * 2 memcpy of 256 item arrays (perhaps after you primop?)
| 
| If the later is not dramatically slower than I
| will bias towards more arrayness. 

I doubt the latter is dramatically slower, but you'd have to experiment
to find out.  And GHC is not doing as well as it should on arrays just
now.  (One of the things on our to-do list.)  Might vary between
implementations too.

Simon