Re: 1-/2-dimensional indexes for common columns, rationale?

>>> create index foo1 on bla (a);
>>> create index foo2 on bla (b);
>>> create index foo3 on bla (a,b);

You say you need faster INSERT performance. Getting rid of some indexes
is a way, but can you tell a bit more about your hardware setup ?
For instance, if you only have one HDD, put an extra HDD in the machine,
and put the database on it, but leave the pg_xlog on the OS's disk. Or the
reverse, depending on which disk is faster, and other factors. Since heavy
INSERTs mean heavy log writing traffic, this almost doubles your write
bandwidth for the cost of a disk. Cheap and efficient. You can also put
the indexes on a third disk, but separating database and log on 2 disks
will give you the most benefits.
If you already have a monster hardware setup, though...