I am affraid this is not something about parameters, but rather missing statistics. Gather statistics on the tables involved and run the queries again with the optimizer_mode on CHOOSE, let us see the plan then.
I would suggest changing one parameter at a time otherwise you are not likely to find out what the problem is. You also need to look at the execution plan before and after aswell as benchmarking your whole application.
It should be noted though in Oracle 8i the db_file_multiblock_read_count directly affects the optimizer in altering the cost of full table scans versus index scans. The higher the value the more likely you are to do full table scans as the optimizer thinks it can read the whole table in less time.
Look at the execution plan with the original settings and then do alter session to change the value. If the execution plan changes from a full table scan to an index scan then my hypothesis is likely to be correct.