The number doesn't represent anything in "real" terms - it's not nanoseconds of delay or anything like that (as it would on CPLDs). It's just a point of comparison: higher speed grades mean better performance. As @austin has said, the datasheet contains minimum specifications for each of the speed grades, so you know that (for example) a -12 device can handle at least 600Mbps DDR across the full temperature range while a -10 can handle at least 500Mbps (the -10 might handle 600Mbps too, but Xilinx makes no guarantee about that).
You can safely generate a bitstream with a lower speed grade. For example, if your design meets timing for a -10 speed grade, you can run that on a -12 chip and it'll definitely work. The reverse is not true; if you take a bitstream that met timing for a -12 device and put it on a -10 device, you may encounter errors.