`produceMessages xs` has bad asyptotic behavior on `length xs`

I had been using code in the form of:

```
xs <- parseFile
produceMessages xs
```

In one case `length xs == 70000` and the process took significant processing power.  Changing the code to instead use `mapM_ produceMessages (chunksOf 1024 xs)` is a huge improvement.  Perhaps the exactly implications of `produceMessages (xs ++ ys)` vs `produceMessages xs >> produceMessages ys` should be documented?
