This is a bit dependent upon your Servive you are using (their demands for quantities) and/or the type of normalization technique used. For our amplicon 16S/18S studies here at the IMR @ Dalhousie (cgeb-imr.ca), we use the SequalPrep PCR purification/normalization kit in 96-well plates (both actions are combined at once in the one kit). The output from these kits is every PCR reaction/sample at about 1-2 ng/uL and same vol. aliquots of the 96 samples are pooled together, meaning the final pool is still at 1-2 ng/uL.

The MiSeq requires an input amplicon library of 4 nM; a pool at ~2 ng/uL of our 578 bp amplicon we sequence (V6-V8 + adaptors + indices) = ~5.4 nM, so we have to dilute slightly to get to the required  4 nM (all quantification needs to be accurate, so using Qbit fluorescence).

So the short answer would be: whatever concentration (in ng/uL) calculates out (given the size of your amplicon) to give you 4 nM (if using MiSeq).

For 454, we used to mix all amplicons to about 20 ng/uL (or a little more) since we/our Service used NanoDrop to quantify and it isn't very reliable below that value. Of course, this pooled library was still way too concentrated for the 454 technique and they wound up diluting it out for the emPCR (but at least we had an accurate starting point).