[AlisdairO]

The code change (and risk) can potentially be quite substantial if you're relying on default SQL server behaviour (i.e. not MVCC/snapshot isolation). Code that relies on blocking when it hits rows that are locked by another process get a bit of a nasty surprise during that kind of switch :-).

[hp50g]

Always have deterministic known behavior :)

We shot ourselves when we first implemented NHibernate by setting our transaction boundary at the wrong place. This caused all sorts of portability problems.