por ahi encontre esto, leelo no seas vago...
Nowhere in the public Microsoft documentation for SQL Server is the unit of measure for the execution plan cost explained. In fact, all Microsoft
SQL Server documentation on this subject seems to be deliberately vague. Is the execution plan cost some measure of time or processor utilization? If it were processor utilization, is it relative to one processor or all processors? An examination of non-parallel and parallel execution plans provide a hint that the unit of measure is most probably time, and that lower plan costs represents lower execution time, but makes no representation of CPU resources.
Parallel execution plans require additional operations that in some cases the additional overhead negates the benefit of a parallel execution plan, but this needs to be determined from actual tests. It has not been determined what the time unit the plan cost represents. A reasonable guess would be that the plan cost is in seconds on a processor that was prevalent at the time SQL Server 7.0 was in development. This could anything from a Pentium 100MHz to a Pentium Pro at 200MHz. This would make sense in that the actual run time was on average 20 times faster on the Xeon 2.4GHz system than represented by the plan cost if the unit of measure was time in seconds.