Query cost refers to the estimated elapsed time, in seconds, required to execute a query on a specific hardware configuration. On other hardware configurations, there is a correlation between cost units and elapsed time, but cost units do not equal seconds. The query governor lets you specify an upper cost limit for a query; a query that exceeds this limit is not run.
Because it is based on estimated query cost rather than actual elapsed time, the query governor does not have any run-time overhead. It also stops long-running queries before they start, rather than running them until they reach a predefined limit.