NotesBench and reality

I was wondering if anybody had actually tested a server on their own and compared to the official Notesbench. Notesbench results do not seem realistic in normal production situations and a fudge factor is usually applied (3 ?).We are contemplating purchasing a DL580 4-way with 3.6 Ghz CPUs for 2,000 users which according to notesbench will allow me 6000 pure iNotes users. The HP sizing tool will give results that seem to be based from the notesbench results and hence recommends DL380 2-way 3.6 Ghz.

Any real life experience with those notesbench results ?

Thanks.