If you mean, how much scaled down version of production to 
be used for performance test environment simulation:

Number of unique users should be the same. I mean if prod-
site gets hit from x uniquie user-ids concurrently at peak 
load, better to simulate same no. of users in your tests( 
which generally is not done so can go with atleast 75%). 
Because no. of users do determine the connection holding 
ability of the servers.

In case of TPS, simulate according to the scalling factor u 
have used to architect your test env against the prod.
Here i mean, if tps rating of prod-h/w divided by tps 
rating of test-h/w is 1.5, then you can try to load your 
test env with full tps load capacity just to ensure test 
systems burn bfore prod-systems burn(if they were to burn 
in case of bad app).