Unmanaged vs. Managed C++ performance

[Nadav]

Hi,

We have a client server archictecture written in C++.
Our client is an agent that runs on thousands of workstations all over
certain organization.
There are thousands of agents that talk to our server on our propriatery
protocol based on TCP.

We have our own infrastructure written in native C++, and this should be
kept because our agent should run on any OS from Win98 and up and with no
installation at all.
Also, our agent should work in lower levels of Windows.
Thus, we would like to keep all of our infrastucture as native C++. All of
the infrastructures modules are built as unmanaged libraries as part of each
of our executables.
Lately, we added a new web application written in C# and a new dll for
interface to a database written in C# as well based on ASP.NET.
Our server inserts data to external database using this dll.
In order to do that, we build our server as managed code and call this dll
explicitly from the server code with #using <Reporter.dll>.

My question is, will we suffer from any interop performance issue in our new
server?
If it's so, where should we have that, between the server to our
infrastructure libraries or between the server to our new dll?
Ofer