I just tried a little experiment myself:
create table foo (
ID int identity,
X float)
set nocount on
declare @ctr int
select @ctr = 0
while @ctr < 1000
BEGIN
insert foo (X)
select rand()
select @ctr= @ctr + 1
END
select X from foo
order by newid()
compute sum(X)
select X from foo
order by newid()
compute sum(X)
drop table foo
The compute clauses in the identical select statements return different results. This is apparently caused by the ordering of the data.