Using NEWID to do a random sort or grab a random number of rows from a result set is a HUGE performance killer and does not scale well. I've had developers slip this kind of stuff into production and in less than a minute the CPUs were pegged at 100%.

Wile this method works, I do not recommend it on anything beyond one time ad-hoc DBA queries or infrequently used applications. You can sort randomly much more efficiently using RAND. Rather than type a lengthy explanation here of how I will submit an article.