With the given information, I'd say you're probably in the ballpark.
The problem of giving an absolute answer is that there are many variables, from the actual PCs that you're testing, to the number and scope of the ACLs in-place between the DMZ and the inside, to the (possible) NAT assignments / maps.
Everything that causes a comparison to occur between the interfaces will occur (potentially, check to see if "turbo" ACLs are available - I don't think they are) for for evey packet that passes between them.
An ACL means that the CPU must examine the packet against every line of the ACL, a NAT or static mapping means that the packet has to be rebuilt (both ways ... think of TCP "acks" as well as bi-di traffic) to change the address and / or port mapping.
Another factor is the load on the PIX. A heavily loaded PIX is likely to operate slower / with more latency than one that is not under heavy load.
All of this plays into the conventions for efficiently setting up your ACLs, filters, address hierarchies & such.
When the system comes under load, every little bit helps to keep it running as fast and efficiently as possible.
Ther are other factors ... the switch(es) and router(s) that feed the PIX, the server(s) that get and send the traffic through the PIX, the quality of the WAN connection .... you get the idea.
Without seeing you configs and a realistic bandwidth analysis on a snapshot of the tested conditions, it's not really possible to give you a fair evaluation.
But, generally speaking, the numbers you present are probably in the zone.
(you did mean MB = MegaBYTES, not BITS, right?)
FWIW
Scott
