Optimization in postgres (like all databases I've used) is super counter-intuitive and baffling at times and the only real way to be sure of anything is to try it yourself on your own data and see (and repeat the test whenever you do a major db upgrade). As an example which may well have changed but definitely was the case the last time I tested (which is circa pg9) it used to be the case that using ANSI join syntax would make a query 5-10% slower. ie

   select a.foo, b.bar
   from a, b
   where a.b_id = b.id 
   and a.something ='blah'

would be reliably 5-10% faster than

   select a.foo, b.bar
   from a
   join b on a.b_id = b.id 
   where a.something ='blah'

even if everything was indexed etc.

Conversely, projecting off unnecessary columns used to result in a big speedup. ie

   select a1.foo, b.bar
   from 
      (select foo, b_id from a where something='blah') a1
   join b on a1.b_id = b.id 

used to be significantly faster than either of those queries on wide tables.

edit: few typos and additional example added