In any case, the query optimizer (provided your rdbms has on) should decide which way to drive the join, based on table statistics and other information in the query.

I remember in one large query the optimizer swapped round the last two tables in the query because it worked out that it needed to read all the rows in the data range on the last table and only half the rows in the other, so swapping them round cut the number of reads required in half. I was quite impressed - it was obvious in hindsight when looking at the data structures, but nobody had noticed...